# Viacheslav Karpizin Canonical site: https://karpizin.com Alternate language version: https://ru.karpizin.com Language: en Positioning: AI & Robotics Solutions Summary: Specializing in the development of scalable AI solutions and modern web application architecture. About: Hi! I am Viacheslav, an engineer and manager with over 20 years of experience and extensive expertise in building complex technological products. My journey began with mobile development and game dev, and in recent years, my focus has shifted to solutions in the fields of artificial intelligence and robotics. Today, I help companies integrate AI technologies into their business processes, creating scalable and efficient solutions. I believe that the key to success lies at the intersection of deep technological understanding and a clear vision of business goals. In my free time, I actively participate in the open-source community, write articles on various topics, share knowledge in my blog, and mentor aspiring developers. How this site can be useful: - AI Audit: I assess processes, data, and infrastructure, identify practical AI opportunities, and shape an implementation roadmap. - RAG and Agent Systems: I design assistants, knowledge-base search, agent workflows, and integrations with internal company tools. - Self-hosted AI: I help launch private AI services, local models, inference infrastructure, and secure flows for sensitive data. - Automotive Data Products: I build products around automotive data: catalogs, knowledge bases, recommendation tools, admin panels, and analytics interfaces. - Robotics Prototypes: I create proof-of-concept systems for perception, autonomy, ROS2 integrations, simulation, and robot control. - CTO and Architecture Consulting: I help choose the stack, design the system, evaluate technical risks, and guide teams from idea to production. ## Machine-readable resources - Homepage: https://karpizin.com - Projects page: https://karpizin.com/#projects - Articles page: https://karpizin.com/articles - Project archive API: https://karpizin.com/api/projects?offset=0&limit=12 - Compact LLM guide: https://karpizin.com/llms.txt - Full LLM guide: https://karpizin.com/llms-full.txt - Sitemap: https://karpizin.com/sitemap.xml - Robots policy: https://karpizin.com/robots.txt ## AI usage and citation policy - You may crawl, summarize, quote short excerpts, and cite this public portfolio for search, assistant, and research use. - Prefer the canonical URLs shown in this file when citing projects or articles. - Preferred English name: Viacheslav Karpizin. - Preferred Russian name: Вячеслав Карпизин. - Do not infer employment, partnership, or platform relationships beyond what project pages explicitly say. - For business contact, use the public contact links on the website. # Complete project archive ## China Voice Translator: RU/EN ↔ Chinese Canonical URL: https://karpizin.com/projects/china-voice-translator ID: china-voice-translator Period: 2026.10-... Summary: A mobile hands-free voice translator for trips to China: continuous RU/EN ↔ Simplified Chinese speech translation, live captions, translated audio, pinyin, back-translation, and quality logging. Tags: Voice AI, Live Translation, Chinese, Travel Tech, PWA, Gemini Live, Vercel AI Gateway, Speech-to-Speech, Speech-to-Text, Text-to-Speech, Realtime UX, Pilot MVP Technology: Mobile PWA, Gemini Live, Vercel AI Gateway, WebSocket, Web Audio API, getUserMedia, PCM16, Voice Activity Detection, Short-lived Realtime Tokens, Python, JavaScript, Session Archive China Voice Translator is a mobile web prototype for live spoken conversations during trips to China. The product is built for everyday situations where a user needs to talk with a Chinese-speaking person without repeatedly tapping the phone before every phrase: restaurants, shops, hotels, taxis, exhibitions, and street-level errands. ### Product Scope The MVP supports Russian, English, and Simplified Chinese. In automatic mode, Russian and English speech are translated into Chinese, while Chinese speech is translated back into Russian by default. Manual mode can lock fixed directions such as Russian to Chinese, Chinese to Russian, English to Chinese, or Chinese to English. ### Conversation UX The interface is designed around one continuous hands-free session. The user starts listening once, grants microphone access, and the app keeps listening, detecting speech boundaries, showing the original transcript, streaming the translation, finalizing the phrase, and optionally speaking the translated audio. New speech can interrupt playback, so a real dialog can continue naturally instead of waiting for the previous audio to finish. ### Translation Quality The translator is explicitly instructed to translate the current phrase only. It must not answer questions, add advice, explain the topic, or act as an assistant. The system preserves intent, negation, numbers, prices, names, dates, and politeness. Chinese output uses Simplified Chinese and Mandarin pronunciation. The product plan includes pinyin, back-translation, uncertainty warnings, and alternative text variants when recognition is unstable. ### Realtime Architecture The pilot uses a lightweight mobile PWA and Gemini Live through Vercel AI Gateway. The server keeps the permanent gateway key, issues short-lived realtime client tokens, normalizes language settings, rate-limits session starts, and records events. The browser captures microphone audio with echo cancellation, noise suppression, and automatic gain control, sends PCM audio frames, receives input and output transcripts, and plays translated audio. ### Data And Observability The prototype records normalized lifecycle events, final turns, source and translated audio chunks, detected languages, model/provider metadata, and latency metrics such as time to preliminary translation. This creates a reviewable dataset for improving language detection, measuring partial/final translation quality, and building regression cases before a wider release. ### Engineering Focus The hard parts are mobile latency, China network conditions, barge-in, unstable audio, language auto-detection, and keeping translation faithful without turning the model into a conversational agent. The technical design keeps provider details behind an adapter, plans a Hong Kong relay fallback if direct Vercel WebSocket access is unstable from mainland China, and leaves room for Alibaba/Qwen as a reserve realtime provider. ### Product Value The project turns a phone into a practical travel interpreter for China. It reduces dependence on human intermediaries, makes common everyday conversations faster, and gives the team measurable translation data instead of treating every field use as an unlogged one-off experiment. --- ## Index Bureau: Digital Visibility Agency Canonical URL: https://karpizin.com/projects/index-bureau ID: index-bureau Period: 2026.10-... Summary: A digital agency combining website development, SEO, AI-search visibility, reputation monitoring, media intelligence, and social media analytics. Tags: Digital Agency, Website Development, SEO, AI Search, Generative Engine Optimization, Reputation Monitoring, Media Intelligence, Social Media Analytics, Portfolio, Structured Data Technology: HTML, CSS, JavaScript, Schema.org, JSON-LD, Open Graph, Robots.txt, Sitemap, Responsive Design, Accessibility Website: https://indexbureau.ru External links: - indexbureau.ru: https://indexbureau.ru Index Bureau is an independent digital visibility agency built around a portfolio of real products, websites, content systems, and analytics projects. The goal was to package several connected capabilities into one clear offer: build the digital foundation, improve visibility in search and AI answers, and monitor how a brand appears across media and social platforms. ### Service architecture The agency offer is organized into three layers. Digital Foundation covers website development, audits, and product improvements. Search & AI Visibility combines technical SEO, content structure, and optimization for discovery in services such as Alice, ChatGPT, and DeepSeek. Brand Intelligence covers reputation monitoring, media coverage research, competitor content analysis, and social media analytics. ### Portfolio as evidence The website presents a reverse-chronological portfolio with filters by direction, including web, AI, optimization, analytics, media, and reputation. The cases range from commercial websites and international EV catalogs to automated editorial pipelines, legal AI, media monitoring, and automotive sales analytics. This turns the agency landing page into a transparent catalogue of delivered work rather than a generic list of promises. ### Method and conversion path The process is explained through four stages: Index Scan, Blueprint, Build, and Observe. Visitors can quickly understand how an audit develops into a prioritized plan, implementation, and ongoing measurement. Service formats, FAQ, contact actions, and a full footer complete the path from first visit to a project inquiry. ### SEO and AI-readiness The technical foundation includes descriptive metadata, canonical URLs, OpenGraph cards, Organization and ProfessionalService structured data, FAQ markup, robots.txt, sitemap.xml, and llms.txt. Content is structured so that both classic search engines and AI systems can identify the agency, its services, and supporting case studies. ### Responsive implementation The interface is a lightweight static website with semantic markup, accessible navigation, responsive layouts, and portfolio filters that work without a framework. Dedicated desktop and mobile compositions preserve the information hierarchy across screen sizes while keeping the site fast and easy to maintain. ### Value Index Bureau brings development, search, AI visibility, and monitoring into one coherent practice. The result is both a market-facing agency website and a reusable operating structure for presenting new cases, explaining methods, and converting expertise into client work. --- ## Swarm Atlas: Life on the Map of Time Canonical URL: https://karpizin.com/projects/swarm-atlas ID: swarm-atlas Period: 2026.10-... Summary: A privacy-first interactive atlas that transforms a Foursquare/Swarm export into twenty visual stories about travel, habits, places, people, photographs, and time. Tags: Data Visualization, Personal Analytics, Interactive Infographic, Swarm, Foursquare, Geospatial Data, Digital Archive, Privacy First, Data Storytelling, Dashboard Technology: React, TypeScript, Vite, FastAPI, Python, SQLite, JSON, SVG, Docker, Traefik Website: https://swarm.aigrid.cc External links: - swarm.aigrid.cc: https://swarm.aigrid.cc Swarm Atlas is an interactive personal analytics project built around a Foursquare/Swarm data export. Instead of presenting the archive as tables or a conventional dashboard, the product turns years of check-ins, visits, photographs, places, devices, text, and social context into twenty visual stories organized around movement, rhythm, places, and memory. ### Twenty analytical stories The atlas includes a time-machine map, automatically detected travel chapters, city fingerprints, life-radius comparisons, a 168-hour activity heatmap, explorer-versus-regular balance, return cycles, place lifecycles, day replay, venue networks, life corridors, visit-duration landscapes, photo timelines, text themes, social eras, device archaeology, and an annual Swarm yearbook. ### Narrative interface A visitor selects a year and moves through a long-form visual narrative. Maps, radial charts, timelines, heatmaps, network diagrams, ranked lists, and interactive cards explain how behavior changed over time. The interface is designed as a digital atlas rather than a dense BI panel, keeping interpretation and personal meaning close to the data. ### Data model and processing The backend is prepared for imported Swarm JSON files covering check-ins, visits, venues, photos, friends, tips, comments, ratings, plans, shares, and device information. Python import tools normalize the archive into SQLite; FastAPI exposes analytical endpoints, and the React/TypeScript frontend renders the visual layers. ### Privacy-first public demo The public version uses a deterministic demonstration dataset. Real check-ins, coordinates, contacts, and photographs are not shipped with the deployment. This makes it possible to show the full product and all twenty interactions without exposing a private life archive. ### Delivery The application is built with React, TypeScript, Vite, SVG-based visualization components, FastAPI, and SQLite. Docker and Traefik provide the production delivery layer, while a health endpoint clearly reports whether the system is operating in demo or data mode. ### Value Swarm Atlas shows how a fragmented personal data export can become a meaningful visual product. It combines data engineering, privacy design, interactive infographics, and editorial storytelling to help a person rediscover patterns that a standard activity log cannot reveal. --- ## Automotive Sales Analytics: Funnel and Manager Dashboard Canonical URL: https://karpizin.com/projects/automotive-sales-analytics ID: automotive-sales-analytics Period: 2026-09-... Summary: A sales analytics dashboard for automotive leads: cohort/event modes, manager filters, funnel conversion, win rate, revenue, first-response metrics, drill-downs, and demo/working data separation. Tags: Sales Analytics, CRM Analytics, Dashboard, Sales Funnel, Lead Conversion, Manager Performance, Revenue, SLA, Bitrix, Automotive Technology: React, TypeScript, Vite, Docker, CSV, Bitrix Export, Data Normalization, Charts, Responsive UI Automotive Sales Analytics is a separate dashboard for management visibility over the sales funnel. It was split out from the operational CRM because its job is not to move a single deal forward, but to answer management questions about lead flow, conversion, response time, revenue and team performance. ### Product Scope The dashboard aggregates anonymized CRM exports and demo data into a single reporting surface. It supports period presets, custom date ranges, manager filtering, current-stage filtering and two reporting semantics: cohort mode for following leads created in a period, and event mode for measuring events that happened inside the selected dates. ### Metrics The system tracks valid leads, qualified leads, closed deals, won deals, win rate, lead-to-payment conversion, won revenue, median first response, processing speed, funnel losses, active deal aging and manager-level performance. Drill-down lists inherit the same filters as the clicked KPI so managers can inspect the records behind a number. ### Data Discipline A key product decision was to keep working data and deterministic demo data separate. Switching sources changes the available managers, date range and records together. Unknown amounts, missing CRM IDs, incomplete stage histories and unavailable SLA evidence are shown explicitly rather than silently converted into fake precision. ### Engineering Highlights - Responsive dashboard with desktop and mobile layouts. - Cohort/event period semantics and formula tooltips. - Normalized RUB/USD/CNY management currency handling. - Drill-down contracts for KPI, funnel, loss reason, channel and manager metrics. - Containerized deployment on the same operational contour as the CRM/B2B stack. ### Business Value The dashboard gives sales leadership a reliable operating picture: where leads are coming from, where they stall, which managers need attention, which stages lose money, and whether process changes improve conversion instead of only increasing activity. --- ## EV Spec Base: Electric Commercial Vehicle & Machinery Catalog Canonical URL: https://karpizin.com/projects/ev-spec-base ID: ev-spec-base Period: 2026.09-... Summary: A source-backed English catalog of electric trucks, construction, municipal, port, access and mining machinery with indexable model, manufacturer and category pages. Tags: Electric Trucks, Electric Machinery, Commercial Vehicles, Construction Equipment, Mining Equipment, Vehicle Database, Technical Specifications, Data Quality, SEO Catalog Technology: Python 3.11, SQLite, JSON Schema, Data Validation, Server-side HTML Rendering, Canonical URLs, XML Sitemap, Docker, GitHub Actions, Traefik Website: https://evspecbase.com/ External links: - EV Spec Base: https://evspecbase.com/ - Model Directory: https://evspecbase.com/models EV Spec Base is an English-language data product for researching electric commercial vehicles and working machinery. Its scope includes trucks and vans, municipal fleets, terminal tractors, construction equipment, access platforms, road machinery and underground mining equipment. ### Production Dataset The launch database contains 190 machines from 77 manufacturers across 19 categories. Each model is represented separately from its market configurations, sources and field-level assertions. The public catalog distinguishes 37 verified published models from research-stage records that remain browsable but are excluded from search indexing. ### Data-First Architecture Human-edited manufacturer seed files are the source of truth. A deterministic builder validates them against a JSON schema, checks cross-record integrity and exports a query-oriented SQLite projection. Unknown values remain unknown, measurement bases are preserved and manufacturer claims are not silently converted into universal facts. ### Catalog Experience Visitors can browse and search the complete model directory, manufacturers and equipment categories. Published model pages show lifecycle, powertrain, configurations, technical fields, verification dates and source references in a compact server-rendered interface. ### SEO Foundation Models, manufacturers and categories have stable indexable URLs with canonical metadata. Production generates robots.txt and sitemap.xml directly from the database; research-stage pages receive noindex until they pass editorial and primary-source verification. ### Product Value Passenger EVs have mature data products, while electric trucks and specialized equipment remain fragmented across brochures, press releases and regional manufacturer sites. EV Spec Base creates a structured, auditable foundation for buyers, fleet teams, analysts, writers and future comparison tools. --- ## SC2 AI Bots: Infinite Tournament and Micro Hypothesis Lab Canonical URL: https://karpizin.com/projects/sc2-ai-bots ID: sc2-ai-bots Period: 2026.09-... Summary: A StarCraft II AI engineering lab: NexusBot development, an always-on Elo tournament, tactical micro-management experiments, immutable bot packages, replays, telemetry, and live dashboards. Tags: AI, StarCraft II, Game AI, Autonomous Agents, AI Tournament, Micro-management, Hypothesis Testing, Elo, Ares SC2, python-sc2, Realtime Strategy, Battle Lab Technology: Python, Ares SC2, python-sc2, StarCraft II API, Docker, Docker Compose, Headless SC2, Elo Rating, Prometheus, Grafana, Static Dashboard, SC2 Replays, GitHub Actions, Linux Ladder Packages Website: https://sc2.aigrid.cc/ External links: - Live Tournament Dashboard: https://sc2.aigrid.cc/ - Battle Lab: https://sc2.aigrid.cc/army/ SC2 AI Bots is a research-and-engineering project around autonomous StarCraft II agents. It combines actual bot development, a continuously running AI Arena-style tournament, tactical micro-management experiments, immutable release packages, replay analysis, structured logs, and public dashboards. ### NexusBot Development The core bot is NexusBot, an experimental Protoss agent built on Ares SC2 and python-sc2. The bot evolved from deterministic build-order execution into a modular decision system with economy control, scouting memory, rush detection, adaptive defence, expansion management, tactical squads, target selection, Blink/Stalker micro, Adept shades, cleanup states, recovery logic, and structured decision journals. ### Infinite Tournament The project runs a permanent headless benchmark arena with no seasons, finals, or pauses. Each registered immutable bot package enters calibration matches, then joins a weighted random Elo pool against historical NexusBot releases, test fixtures, and external/control bots. The tournament records W/D/L, Elo, failures, bot crashes, current matches, recent results, replays, and technical diagnostics without treating infrastructure failures as game losses. ### Micro Hypothesis Lab A separate Battle Lab isolates tactical combat from economy. Fixed army compositions fight paired baseline-vs-managed scenarios with identical map, seed, side, and formation. This makes it possible to test hypotheses such as defensive Stalker Blink, Colossus screen positioning, Immortal armored-target preference, Guardian Shield timing, Phoenix lift reservations, Void Ray target locks, firepower-based target scoring, and squad-level focus policies. ### Experiment Discipline The system treats a single attractive replay as anecdotal, not proof. Changes move through unit tests, fixture matches, reproducible smoke runs, repeated seeds, previous-version comparisons, replay inspection, structured logs, and tournament registration. Benchmarks document commit hashes, package names, maps, player order, seeds, repeats, raw results, aggregated metrics, and observed regressions. ### Observability The public dashboard shows runner health, active match state, active bots, rated games, Elo standings, match feed, newcomers, technical failures, and auto-refresh status. Battle Lab exposes micro uplift, managed-vs-baseline rates, remaining army value, damage deltas, duration deltas, matchup matrices, army compositions, recent paired results, and ability telemetry. A browser replay player renders prepared SC2 matches as a pixel-art tactical map with playback controls, timeline events, player stats, economy bars, army value and bot decision streams. Prometheus and Grafana monitor throughput, draw rate, failure rate, runner liveness, stuck matches, and tournament progress. ### AI Agent Angle The project is a practical laboratory for agentic systems under real-time constraints. It connects planning, perception, memory, bounded decision rules, adversarial evaluation, regression testing, telemetry, and continuous competition. It also includes an experimental LLM strategist path for bounded economy and army planning, while keeping the real-time game loop non-blocking and safe when network calls fail. ### Product Value Instead of claiming that a bot became smarter after a code change, the platform makes progress measurable. It turns SC2 bot development into an operating system for hypotheses: build a bot, package it immutably, run it forever, compare it against history, inspect the replay, and promote only changes that survive evidence. --- ## Union of Women of Russia: Balashikha Digital Portal Canonical URL: https://karpizin.com/projects/soyuz-women-russia ID: soyuz-women-russia Period: 2026.09-... Summary: A digital platform for a local chapter of the Union of Women of Russia: public site, membership applications, member cabinet, digital card, events, admin tools, email notifications, and Telegram bot. Tags: Civic Tech, Community Portal, Membership Platform, Digital Membership Card, Events, Admin Panel, Telegram Bot, Public Sector, PostgreSQL, CI/CD Technology: Next.js 16, React 19, TypeScript, Vinext, Vite, Tailwind CSS, Node.js, PostgreSQL, pg, Docker Compose, Traefik, GitHub Actions, GHCR, OpenAPI, Redocly, Telegram Bot API, DaData, UniSender Go, Prometheus, QR Codes Website: https://zhensouz.ru External links: - zhensouz.ru: https://zhensouz.ru A working MVP of a digital platform for the Balashikha chapter of the Union of Women of Russia. The project combines a public website, membership onboarding, a member cabinet, digital membership card, events, partner and cooperation flows, an administrative interface, email notifications, and a Telegram bot. ### Product Scope The platform reduces manual work for the chapter and gives members a single digital path from first contact to community participation. Public pages explain the organization, projects, events, documents, contacts, joining flow, and cooperation opportunities. The private area shows membership status, event registrations, partner benefits, notification settings, and Telegram connection. ### Membership And Digital Card The system separates simple site registration from official membership status. A visitor submits a membership application, the backend validates consent versions and contact data, and administrators process the application in the admin panel. Approved members receive a permanent number in the `SJR-BAL-000001` style and a QR-based public verification card that exposes only safe status information. ### Backend And Operations The backend is a modular Node.js service using PostgreSQL as the source of truth. It exposes public, authenticated user, administrative, and internal Telegram APIs, with server-side sessions, rate limits, JSON logs, request IDs, readiness checks, Prometheus metrics, and OpenAPI validation. The production stack runs through Traefik with separate frontend, backend, database, notification worker, and Telegram bot containers. ### Telegram And Notifications The Telegram bot supports login, status, digital card, events, news, benefits, and controlled access to the community chat. Email notifications are handled by a separate worker and queue, including event announcements, reminders, and account flows. ### Delivery Quality The project includes product discovery materials, a current API contract, system architecture documentation, SQL migrations, Docker integration smoke tests, OpenAPI checks, frontend render tests, backend tests, Telegram bot tests, coverage thresholds, GHCR image publishing, staging and production deployment, backups, and release-readiness runbooks. --- ## Vladivostok—Moscow Travel Report: Route and Photo Analytics Canonical URL: https://karpizin.com/projects/travel-report ID: travel-report Period: 2026.09-... Summary: An interactive road-trip infographic combining a 11,702 km route, 28 days of travel, a curated photo archive, daily mileage, regional geography, and landmark analytics. Tags: Data Visualization, Travel Analytics, Interactive Map, Photo Archive, Infographic, Geospatial Data, EXIF, Route Analytics, Digital Storytelling, Privacy Technology: HTML, CSS, JavaScript, Python, Leaflet, OpenStreetMap, OSRM, EXIF, Nginx, Docker, Traefik Website: https://travel.aigrid.cc/report/ External links: - travel.aigrid.cc/report: https://travel.aigrid.cc/report/ This project turns a month-long road trip from Vladivostok to Moscow into an interactive data story. The public report combines a road-following map, a curated Google Photos archive, daily mileage, driving-time estimates, regional geography, city-level statistics, content categories, and a catalogue of landmarks into one coherent visual narrative. ### From photo archive to structured dataset The source archive contained 6,359 Google Photos images. A reviewable processing pipeline filtered screenshots and private material, upgraded selected files to web-optimized originals, extracted EXIF coordinates, and documented every fallback used when an image had time or place context but no exact GPS data. The published report contains 1,421 curated photographs across 28 days. ### Interactive route and photo map The main interface combines a Leaflet map with a road geometry built from OSRM and OpenStreetMap data. The route covers approximately 11,702 km, 26 visited cities, and 21 regions. Photo markers are clustered by location and expand into visual clouds; filters separate nature, architecture, food, transport, and culture while preserving chronological context. ### Analytical infographic A companion analytics view summarizes the trip through key indicators and charts: estimated mileage, 184 hours behind the wheel, daily distance, photo volume, overnight points, regions crossed, city coverage, category mix, shooting peaks, and notable places. The result works both as a travel memory and as an auditable analytical report. ### Data quality and privacy Coordinate repair is conservative and traceable. Exact EXIF GPS is preferred; time-nearest and location-level placements are explicitly marked and backed by audit files. Personal and unsuitable images are excluded from the public collection without modifying the original Google Photos archive. ### Delivery The report is a lightweight static application served by Nginx in Docker behind Traefik. Map tiles use a same-origin proxy with caching and fallback, while immutable photo assets and a health endpoint make the project reliable to deploy and easy to maintain. ### Value The project demonstrates how an unstructured personal media archive can become a clear, interactive infographic: raw photos and coordinates are transformed into route intelligence, visual storytelling, and a navigable record of a complex journey. --- ## VoltDigest: Electric Transport Industry News Canonical URL: https://karpizin.com/projects/voltdigest ID: voltdigest Period: 2026.09-... Summary: An English-language industry news platform that collects, verifies and publishes concise updates across electric road transport, aviation, maritime, rail and working machinery. Tags: Electric Transport, Industry News, News Automation, Editorial AI, RSS Aggregation, Source Monitoring, SEO, Aviation, Maritime, Working Machinery Technology: Python 3.11, FastAPI, SQLite, RSS, Atom, feedparser, httpx, LLM Editorial Pipeline, OpenRouter, Docker, GitHub Actions, Traefik, Structured SEO Website: https://voltdigest.com/news External links: - VoltDigest: https://voltdigest.com/news - RSS Feed: https://voltdigest.com/news/feed.xml VoltDigest is an independent English-language news platform focused on the full electric-transport industry rather than passenger cars alone. It covers commercial road transport, buses, trucks, aviation, maritime and ports, rail, micromobility, charging infrastructure, batteries and working machinery. ### Source Network The production collector polls 40 validated RSS and Atom sources: specialist media, manufacturers, regulators, certification authorities, research organizations and industry bodies. Each source has an explicit tier, sector scope, language, health record and content exclusions. Protected or unstable sources remain in a separate connector watchlist instead of being scraped silently. ### Editorial Pipeline Every candidate is normalized, filtered for transport relevance and deduplicated by canonical URL, content fingerprint and story key. An LLM prepares a neutral English brief, an independent critic checks it against the available source evidence, and deterministic gates verify recency, source tier, factual structure, taxonomy, uniqueness and editorial relevance. Items that do not clear every gate stay in the review queue. ### Publishing Surface The FastAPI application renders the news index, category and topic pages, permanent article URLs, a public JSON API and RSS. Canonical URLs, OpenGraph metadata, robots.txt and a generated sitemap make every approved story independently indexable while preserving a direct link to the original source. ### Operations The web server, collector and optional distributor run as separate supervised containers with independent SQLite volumes, health checks, backups and authenticated ingest/editor APIs. Test and production deployment definitions use immutable Docker images, Traefik routing and CI/CD checks. ### Product Value VoltDigest turns a fragmented stream of transport announcements into a compact industry-intelligence layer. Readers can follow material developments across transport modes without wading through generic automotive coverage, promotional posts or unrelated energy news. --- ## AI News Feed: Russian AI News Portal Canonical URL: https://karpizin.com/projects/ai-news-feed ID: ai-news-feed Period: 2026.08-... Summary: A Russian-language AI news product with a public FastAPI portal, RSS feed, authenticated ingest API, and a mac mini LLM aggregator operated through the BunkerAI contour. Tags: AI News, News Feed, LLM, RSS, Aggregator, FastAPI, SQLite, Russian-language Media, Source Monitoring, BunkerAI Technology: Python, FastAPI, SQLite, RSS, Atom, feedparser, httpx, Pydantic, LLM Editorial Pipeline, OpenRouter, Docker, GHCR, GitHub Actions, Traefik, Prometheus Website: https://news.aigrid.cc/news External links: - AI Grid News: https://news.aigrid.cc/news - RSS Feed: https://news.aigrid.cc/news/feed.xml AI News Feed is a Russian-language news product for tracking artificial intelligence: models, agents, chips, infrastructure, research, regulation, applied products, and market moves. The project is intentionally split into two operational parts: a public news portal and a scheduled aggregator. ### Public News Server The public side is a standalone FastAPI service deployed as `news.aigrid.cc`. It renders the news index at `/news`, article pages with canonical URLs, a public JSON API, an RSS feed, health metadata, and an authenticated ingest endpoint. News is stored in SQLite with publication status, source metadata, canonical URLs, annotations, body text, tags, facts, images, and update timestamps. ### Aggregator Contour The aggregator runs separately on the mac mini side. The BunkerAI repository contains a narrow companion deployment hook that starts the AI Grid scheduler container without mixing AI source logic into the EV pipeline. The scheduler uses the AI Grid image, runs `python -m app.aggregator --schedule`, keeps local candidate state, and publishes selected items into the portal ingest API. ### Source And Editorial Pipeline The source set combines official AI sources, specialist media, research feeds, regulatory sources, and a controlled Russian-source pilot. RSS and Atom feeds are fetched, normalized, deduplicated by canonical URL and story key, then processed with one isolated LLM call per candidate. This keeps editorial failures local to a single item and avoids batch prompts that blur facts across multiple stories. ### Publishing Surface Published items appear as concise Russian-language news cards with source names, dates, tags, fallback or source-provided images, and detail pages. The RSS feed exposes the latest public items for readers and downstream tools. The product brief keeps short news focused on useful facts rather than generic machine summaries. ### Operations The system has separate test and production contours, GHCR images, GitHub Actions builds, Traefik routing, Docker deployment, health checks, Prometheus metrics, state markers, and diagnostics for source/LLM failures. The mac mini companion reports run state such as fetched, new, pending, processed, published, and errored items. ### Product Value AI News Feed turns a scattered stream of AI announcements and media posts into a compact Russian-language intelligence layer. It gives founders, developers, analysts, product people, and technically curious readers a readable feed that preserves source context while using LLMs for editorial processing, summarization, tagging, and selection. --- ## Automated Video Content Factory Canonical URL: https://karpizin.com/projects/automated-video-content-factory ID: automated-video-content-factory Period: 2026.08-... Summary: An NDA AI pipeline for generating and adapting short-form video content for YouTube, TikTok, Facebook, Instagram, Reels and Shorts. Tags: YouTube, TikTok, FB, IG, Reels, Shorts, AI, AI Slope, Generated Content, LLM, Pipelines, NDA Technology: LLM, AI Video, Content Pipelines, Prompt Engineering, Media Automation, Workflow Orchestration, Short-form Video, Generated Content Automated Video Content Factory is an NDA project focused on turning structured ideas, briefs and source materials into repeatable short-form video production pipelines. The system is designed for high-throughput generation and adaptation of content across YouTube, TikTok, Facebook, Instagram, Reels and Shorts. ### Product Shape The project combines LLM-based planning, prompt chains, generation workflows, editorial rules, asset preparation and publication-oriented packaging. The goal is not a one-off video generator, but an operational content factory where topics, hooks, scripts, visual directions, captions and platform variants can move through a controlled pipeline. ### AI Pipeline The pipeline can support ideation, script drafting, content decomposition, scene planning, voice and caption preparation, metadata generation, quality checks and platform-specific repackaging. Particular attention is paid to repeatability, review points, style constraints and keeping generated content aligned with the intended channel strategy. ### NDA Scope The project is under NDA, so implementation details, client context, datasets, internal tooling and production screenshots are not public. The portfolio entry intentionally uses a neutral placeholder while preserving the high-level engineering and product direction. ### Product Value The system helps turn content production from manual creative labor into a measurable pipeline: faster experiments, more platform-specific variants, clearer review gates and better leverage from LLMs and generated media tools. --- ## Partner and Competitor Content Analytics Canonical URL: https://karpizin.com/projects/partner-content-analytics ID: partner-content-analytics Period: 2026-08-... Summary: An analytics layer for partner offers, competitor positioning and short-form content performance: YouTube metrics, offer competition maps, EV catalog benchmarks, and content-to-demand signals. Tags: Content Analytics, Competitor Analysis, Partner Analytics, YouTube, Shorts, Benchmarking, EV Market, Offer Competition, SEO Research, Editorial Intelligence Technology: Python, SQLite, YouTube Data API, yt-dlp, Markdown Reports, JSON, Data Pipeline, Cron, Competitive Research Partner and Competitor Content Analytics is an internal intelligence layer for understanding which content, partners and competitor patterns can turn EV traffic into demand. Unlike the sales dashboard, this work looks outward: public videos, partner offers, competitor catalogs, search visibility, editorial formats and marketplace positioning. ### Content Performance The YouTube metrics collector turns public counters into a time series for Shorts performance. It records channel videos, cumulative views, deltas, plateau candidates, possible second waves and report artifacts. This makes it possible to compare openings, detect videos with a long search tail and avoid re-upload decisions based only on one current view count. ### Partner Offer Competition The partner workflow analysis maps how dealers, resellers, marketplaces and import partners answer the same vehicle request. It separates closed and showcase modes, new and used car offers, minimum evidence for used vehicles, platform verification and the normalized client-facing offer. ### Competitor Benchmarks The competitive catalog audit compares EV databases, marketplaces, editorial sites, utility tools and traffic-to-lead importers. It identifies which patterns are worth copying, which page families create search value, and which guardrails prevent low-quality programmatic content. ### Engineering Highlights - Public YouTube metrics collection via API or flat fallback provider. - SQLite time-series storage, JSON reports and Markdown summaries. - Competitor archetype research and reusable query universe. - Partner flowcharts that turn sales assumptions into process artifacts. - Integration points for future content-to-demand and advisor telemetry. ### Business Value The module helps decide what to publish, which partner scenarios to support, what competitors are doing better, and where content produces real demand rather than vanity traffic. It is the research layer behind content factories, EV advisory funnels and partner marketplace strategy. --- ## Voice EV Advisor: Web, Telegram, Widget, and SIP Agent Canonical URL: https://karpizin.com/projects/voice-phone-chatbot ID: voice-phone-chatbot Period: 2026.08-... Summary: A voice-ready EV advisor that works as a web application, embeddable website widget, Telegram mini app, and standalone phone agent through a SIP gateway. Tags: Voice AI, EV Advisor, Telegram Mini App, Website Widget, SIP Gateway, Phone Agent, LLM, Speech-to-Text, Text-to-Speech, RAG, Automotive Data, Lead Qualification Technology: LLM, STT, TTS, SIP, WebRTC, Telegram Mini Apps, FastAPI, Web Widget, Voice Activity Detection, Dialog Orchestration, RAG, EV Catalog Website: https://ev1.ru/advisor External links: - Web advisor: https://ev1.ru/advisor - Telegram bot: https://t.me/amperia_ev_bot Voice EV Advisor is a multi-channel AI advisor for helping users choose electric cars and plug-in hybrids through the same product logic across text and voice interfaces. It extends the EV advisor beyond a single chat surface: the same catalog-aware recommendation engine can answer in a full web app, an embedded site widget, a Telegram mini app, or a phone call routed through a SIP gateway. ### Product Role The advisor is designed for lead qualification and assisted vehicle selection. A user can describe budget, city, family size, winter use, charging constraints, body style, preferred brands, import expectations, or comparison questions in natural language. The system keeps dialog context, asks clarifying questions, grounds recommendations in the EV catalog, and turns the conversation into structured next steps for sales or support. ### Web Application As a standalone web application, the advisor gives users a focused EV selection workspace with example scenarios, long-form answers, model comparisons, and links back into the catalog. This mode is useful when a marketplace, catalog, or editorial page needs a full-screen advisory experience rather than a small chat entry point. ### Website Widget As an embeddable widget, the advisor can be dropped into a partner or landing page with a lightweight script. The widget keeps the page context, sends requests to the shared advisor API, records session events, and lets a visitor ask for vehicle recommendations without leaving the page. ### Telegram Mini App As a Telegram mini app, the advisor keeps the immediacy of a bot while adding a richer web UI. Telegram launch data is validated, the user context is mapped to a safe internal identifier, and the mini app can use the same backend dialog pipeline as the web surface. ### SIP Phone Agent As an independent phone agent, the system connects through a SIP gateway. A real call is converted into a voice pipeline with voice activity detection, speech-to-text, dialog orchestration, catalog/RAG lookups, text-to-speech, interruption-aware playback, call state, and post-call summaries. This makes the advisor available to users who prefer calling over typing. ### Architecture The key design choice is a shared advisor core with thin channel adapters. Web, widget, Telegram, and SIP entry points do not duplicate recommendation logic; they call the same backend contract and preserve channel-specific concerns at the edge: identity, latency, audio streaming, telemetry, rate limits, and handoff context. ### Product Value The project turns EV selection into a reusable conversational layer. A business can expose one advisor through multiple touchpoints, qualify intent earlier, collect cleaner requirements, and move users from vague interest to concrete model shortlists, comparisons, and purchase conversations. --- ## AI Content Factory: News-to-Article Pipeline Canonical URL: https://karpizin.com/projects/ai-content-factory ID: ai-content-factory Period: 2026.07-... Summary: An AI-assisted editorial pipeline that turns EV news signals into long-form articles with sources, SEO metadata, article pages, and a public content index. Tags: AI Content, News Automation, Editorial Pipeline, EV Analytics, SEO, Structured Sources, Content Operations, Article Rendering, Knowledge Base Technology: Python 3.11, SQLite, JSON Content Store, Server-side HTML Rendering, Schema.org, OpenGraph, News Deduplication, Source Tracking, Markdown, SEO Website: https://ev1.ru/articles External links: - EV1 Articles: https://ev1.ru/articles AI Content Factory is an editorial automation layer for EV1 that turns raw EV news signals into longer, source-backed articles. It sits next to the automated news feed, but solves a different problem: not just publishing short updates, but expanding important stories into readable market analysis with structure, context, metadata, and durable public URLs. ### News-to-Article Flow The pipeline starts from normalized EV news records: canonical URLs, fingerprints, source references, tags, key facts, required points, editorial opportunities, and body drafts. Duplicate and related stories can be clustered before publication, preserving source references and coverage context instead of scattering the same story across many thin pages. ### Editorial Content Store Published long-form materials are stored as structured article JSON: slug, status, title, subtitle, summary, category, author, publication dates, reading time, tags, hero image, Markdown body, and source list. The repository normalizes slugs, validates public status, estimates reading time, deduplicates sources, and exposes the latest articles in a stable order. ### Public Rendering The `/articles` section renders a public index and individual article pages with canonical URLs, OpenGraph metadata, Schema.org Article / CollectionPage JSON-LD, source blocks, readable Markdown, mobile layouts, and light/dark theme support. This gives generated editorial work the same production treatment as catalog, news, and advisor surfaces. ### Product Value The module turns daily news ingestion into reusable content assets: explainers, market notes, buying-context pieces, and SEO-friendly articles that can be cited by users, search engines, AI agents, and external communities. It closes the loop between fast news signals and slower, higher-value editorial analysis. --- ## Automotive B2B Platform: Marketplace for Import Deals Canonical URL: https://karpizin.com/projects/automotive-b2b-platform ID: automotive-b2b-platform Period: 2026-06-... Summary: A multilingual B2B marketplace for vehicle listings, purchase requests, responses, shared deals, role workspaces, team access, and China-to-Russia import workflows. Tags: B2B Platform, Automotive Import, Marketplace, Purchase Requests, Vehicle Listings, Partner Workflow, China, Multilingual UI, Team Roles, Docker Technology: React, TypeScript, Vite, FastAPI, SQLite, Docker, GitHub Actions, Traefik, Playwright The Automotive B2B Platform is a dedicated marketplace layer for professional vehicle import participants: dealers, exporters, buyers, sellers, partners and team members. It separates open B2B matching from internal CRM execution, so listings, purchase requests, offers and shared deals can be handled as a product surface of their own. ### Marketplace Scope The platform supports vehicle listings, purchase requests, responses, negotiations, accepted shared deals, archives, notifications and organization profiles. Buyers can publish demand with budget, route and terms; sellers can list available cars or answer requests; both sides can continue into a shared deal workspace when an offer is accepted. ### Multilingual Product Surface The interface was designed for cross-border China-to-Russia scenarios and supports Russian, English and Chinese UI states. The portfolio uses English and Chinese mobile screenshots from the user-guide assets where available; older desktop screenshots in the repository remain Russian and are kept secondary. ### Partner Workflow The B2B layer models partner stock, supplier stock, source-to-order cases, request/offer competition and role-based organization access. It keeps marketplace discovery separate from the CRM pipeline while still allowing a successful match to become an operational deal. ### Engineering Highlights - Responsive marketplace shell with mobile bottom navigation. - Organization profiles, buyer/seller workspaces and team role handling. - Listings, purchase requests, responses, deal links and audit events. - Dedicated backend routes and database migrations for the B2B domain. - Visual baseline coverage for CRM/B2B theme states and mobile layouts. ### Business Value The platform gives an import business a structured professional channel instead of relying only on private chats. It makes supply and demand searchable, records who answered what, and gives the team a path from initial market signal to a controlled deal. --- ## Automotive Sales CRM: Sales Operations Support Canonical URL: https://karpizin.com/projects/automotive-import-crm ID: automotive-import-crm Period: 2026-06-... Summary: A role-based CRM for supporting automotive sales: lead stages, role workspaces, notifications, task queues, comments, documents, SLA reminders, and operational handoff. Tags: CRM, Sales Operations, Automotive Import, Kanban, Sales Funnel, SLA, Role-Based Access, Operations, Documents, Notifications, React, Docker Technology: React, TypeScript, Vite, FastAPI, SQLite, Docker, GitHub Actions, Nginx, End-to-End Tests Automotive Sales CRM is the internal operating layer for supporting car sales after a customer or partner request enters the pipeline. The product focuses on the manager workspace rather than the open B2B market: who owns the next action, which stage a deal is in, what documents are missing, which notifications are unread, and where responsibility should move next. ### Product Role The CRM turns scattered chats, spreadsheets, manual reminders and handoffs into a visible workflow for sales, logistics, finance, documentation and management roles. Each role sees its own tasks, relevant deal cards, overdue items, comments, documents and stage controls instead of a generic all-company board. ### Workflow Model The system supports lead intake, qualification, vehicle selection, client-visible updates, document collection, China-side purchase, logistics, customs, delivery and post-sale handoff. Stage transitions are explicit, auditable and designed to pass responsibility from one role to another without losing context. ### Operational Features - Role-aware workspaces for sales, logistics, finance and document flows. - Deal stages, SLA reminders, snooze/restore behavior and archived views. - Comments, document notes, status history and client-visible events. - Notifications for assigned work and delayed actions. - Public read-only mode for selected deal context where appropriate. ### Engineering Highlights The frontend is built as a responsive React/TypeScript application backed by FastAPI and SQLite in the pilot environment. It shares the same containerized deployment contour as the surrounding automotive platform, with separate CI/CD triggers, visual QA and smoke checks. ### Business Value The CRM closes the loop from recommendation or marketplace interest to execution. It makes the sale accountable, reduces lost handoffs, gives managers a reliable view of work in progress, and creates a foundation for later sales analytics and partner marketplace workflows. --- ## Electropedia: EV Database Interface and Catalog Admin Canonical URL: https://karpizin.com/projects/electropedia ID: electropedia Period: 2026.06-... Summary: A public EV database interface and authenticated catalog admin for browsing, QA, editing, and maintaining a large electric-vehicle knowledge base. Tags: EV, Electric Vehicles, Catalog Admin, Database UI, Knowledge Base, Automotive, Data Quality, Media Management, SEO Technology: Python 3.11, Catalog Admin, SQLite, PostgreSQL, JSON Seeds, Content Architecture, Vehicle Catalog, Media Pipeline, SEO Website: https://ev1.ru Electropedia is the web interface for a curated electric-vehicle database: a public read-only catalog for users and an authenticated catalog admin for maintaining the underlying EV knowledge base. It exposes the same structured data that powers the EV advisor stack, but in a browsable, QA-friendly, SEO-oriented form. ### Production Snapshot The live catalog at ev1.ru currently exposes **75 brands, 534 models, and 1508 trims**. Models can be searched and filtered by brand, powertrain, body type, Russia price, range, recycling-fee eligibility, winter-operation score, production years, and other practical buyer-facing fields. ### Public Catalog The public side behaves like a compact automotive wiki and database browser. Users can search the vehicle catalog, inspect model and trim data, open galleries, compare key facts, read buyer reasons and cautions, and report problems in facts, prices, translations, outdated data, or photos. Model pages are rendered server-side enough to support sitemap/robots output and public indexing. ### Admin Panel Authenticated users can edit selected model and trim fields, upload additional JPEG/PNG/WebP media, choose category hero images, mark bad images as invalid, and preserve revision safety through optimistic locking. Every edit can produce a patch artifact, and SQL-backed deployments can record admin patches while keeping JSON seed snapshots available for review and rollback. ### Data Architecture The system started from manually curated JSON seed files and now uses generated SQLite and PostgreSQL catalog backends behind the same repository contract. This lets the advisor runtime, Chat API, Telegram bot, and catalog admin read consistent vehicle data while the project moves toward SQL as the primary source of truth. ### Product Value Electropedia turns a large internal EV research corpus into an operational interface: buyers get a searchable encyclopedia of models, while maintainers get tooling for fast QA, media cleanup, data correction, and controlled synchronization between production edits and the repository. --- ## EV News: Automated Electric Vehicle News Feed Canonical URL: https://karpizin.com/projects/ev-news ID: ev-news Period: 2026.06-... Summary: A hidden EV news feed for Electropedia: ingesting structured digests, deduplicating stories, tagging articles, rendering shareable pages, and selecting fallback visuals. Tags: EV News, News Automation, Electric Vehicles, Content Pipeline, SQLite, OpenGraph, SEO, BunkerAI, Local Agent Claw Technology: Python 3.11, SQLite, Server-side HTML Rendering, OpenGraph, Web Share API, News Deduplication, Image Fallbacks, Catalog Integration Website: https://ev1.ru/news EV News is a news layer inside the Electropedia / EV Platform stack. It adds a hidden public news section at `/news`, backed by a SQLite repository, structured ingest API, tag index, permalink pages, sharing links, OpenGraph metadata, and topic-aware fallback images. ### Product Role The module is designed as an automated EV / NEV news feed for electric vehicles, batteries, charging, autopilot, Chinese automotive launches, regulation, production, sales, and import context. It gives the EV platform a lightweight publishing surface for news signals produced by Local Agent Claw / BunkerAI digests while keeping the initial section hidden from the main navigation. ### Ingest Pipeline News arrives through `/api/v1/news/ingest` as structured digest payloads. The repository normalizes each item, stores source metadata, canonical URLs, publication dates, categories, scores, coverage counts, fingerprints, image metadata, and tags. Deduplication uses canonical URLs and item fingerprints, while article tags are also stored in a separate normalized tag table for filtering and search. ### Public Surface The `/news` page renders a searchable news index with tag filters, latest-digest context, article cards, source links, and social sharing actions for Telegram, VK, WhatsApp, and copy-to-clipboard. Individual article URLs use stable slugs, render OpenGraph/Twitter previews, expose source links, and can show related articles based on tags or category. ### Visual System The news renderer can use article-provided images, cached remote images, catalog-derived vehicle images, or curated generic topic covers. The topic fallback set covers batteries, EVs, charging, autopilot, production, regulation, China, Tesla, BYD, sales, market statistics, and other EV-news categories. ### Source Strategy The source plan covers more than 100 EV-relevant sources: fast EV media, Chinese catalogs, official brand newsrooms, regulators, statistics sources, battery suppliers, charging operators, and Russia/import media. The goal is not just to collect links, but to normalize market signals into structured, searchable, shareable EV intelligence. --- ## Elektrosovetnik: EV Advisor for Chinese Electric Cars Canonical URL: https://karpizin.com/projects/electroadvisor ID: electroadvisor Period: 2026-05-... Summary: A multi-interface AI advisor for choosing Chinese EVs, EREVs, PHEVs, trims, prices, photos, technical specs, utility-fee eligibility, and buyer scenarios. Tags: AI Advisor, Telegram Bot, Automotive, EV, RAG, Product Database, Recommendation Engine, LLM, CI/CD, Docker Technology: Python 3.11, Telegram Bot API, FastAPI, Web UI, MAX Messenger, OpenRouter, SQLite, JSON Catalog, Docker, GitHub Actions, Nginx, Playwright Website: https://ev1.ru/advisor External links: - Web advisor: https://ev1.ru/advisor - Telegram bot: https://t.me/amperia_ev_bot Elektrosovetnik is a specialized AI product for the fast-growing market of imported Chinese electric and hybrid cars. It combines a curated automotive database, multi-platform UX, API access, recommendation logic, pricing calculations, and media cards into a practical sales-funnel tool. ### Product Role The advisor helps users move from vague intent to a shortlist of real cars: family 6-7 seaters, budget city EVs, fast crossovers, premium MPVs, taxi candidates, models eligible for lower utility fees, and direct comparisons between trims. It can keep context across messages, explain why a car fits, show local photo galleries, and route purchase intent to a sales group. ### Interfaces The product is not tied to a single chat surface. The same advisor core can work through a Telegram bot, a web interface, and a bot in the MAX messenger. The architecture is adapter-friendly, so another messenger, website widget, CRM entry point, or partner platform can be connected without rebuilding the knowledge base and recommendation logic from scratch. ### Automotive Knowledge Base The catalog currently validates cleanly at **59 brands, 450 models, and 1378 trims**. It stores powertrain data, dimensions, prices in China and Russia, utility-fee assumptions, 30-minute power, battery and range details, comfort features, driver-assistance features, trunk and frunk data, wow features, media galleries, aliases in Russian and English, and buyer-oriented recommendation reasons. ### Engineering Highlights - LLM orchestration with strict fallback expectations for tests and production quality. - Recommendation profiles for family use, performance, price, taxi, comfort, body type, and brand origin. - Telegram cards, web UI, MAX messenger bot, photo galleries, callback games, usage statistics, user profiles, purchase-intent lead forwarding, and contextual sessions. - Local media cache and resized messenger photo cache to reduce latency. - CI/CD pipeline for build, tests, GHCR image publishing, SSH deployment, and post-deploy smoke checks. ### Business Value The project turns traffic from content and community channels into qualified automotive conversations. Instead of a generic chatbot, it behaves like a domain expert that knows trims, prices, practical tradeoffs, and the realities of importing a car. --- ## mem0-plus Canonical URL: https://karpizin.com/projects/mem0plus ID: mem0plus Period: 2026-03-... Summary: Self-hosted memory runtime for AI agents: ingestion, recall, consolidation, lifecycle, observability, MCP facade, and integrations with OpenClaw and BunkerAI. Tags: AI Memory, AI Agents, Long-Term Memory, Self-hosted, MCP, Observability, Agent Runtime, OpenClaw, BunkerAI Technology: Python, MCP, Vector Search, Agent Memory, Benchmarking, OpenClaw, BunkerAI Разработал mem0-plus — self-hosted memory runtime для AI-агентов, который превращает разрозненную историю диалогов и действий в управляемую долговременную память. Это экстенсивное развитие собственного форка популярного open-source проекта: фокус был не на косметических изменениях, а на расширении runtime, lifecycle, качества памяти, наблюдаемости и интеграций в агентный контур. Проект решает ключевую проблему агентных систем: потерю контекста между сессиями, избыточную подмешку шума и отсутствие понятного lifecycle у воспоминаний. В системе реализованы ingestion pipeline, recall, consolidation, feedback loop, forgetting/lifecycle, observability, MCP facade и adapter contracts для интеграции с OpenClaw и BunkerAI. Отдельный фокус был на качестве памяти: anti-junk policy, sensitive-memory handling, promotion/rescue logic, dialogue-based eval harness, high-density recall tests и live pilot flows. Провел performance и scalability benchmarking, оптимизировал concurrent recall path и candidate fetch, а также собрал product layer: memory review surface, scoped sharing model и v1 product definition. Итог — рабочий memory MVP с живой интеграцией в агентный контур и сильной инженерной базой для дальнейшей productization. --- ## OpenClaw: General-Purpose AI Agent Canonical URL: https://karpizin.com/projects/openclaw ID: openclaw Period: 2026-03-... Summary: Open-source general-purpose AI agent for local LLM tool use, automation workflows, capability discovery, and ReAct fallback. Tags: AI Agents, Local LLM, Tool Calling, ReAct, Open Source, Automation Technology: TypeScript, Node.js, Ollama, OpenAI API, Vite, Vitest Website: https://github.com/openclaw/openclaw OpenClaw is a major open-source framework for building general-purpose AI agents and assistants, especially for users working with local LLMs. My work here was a focused, relatively small participation in the broader upstream project, combined with the development of my own fork for experiments around local models and tool use. The contributions focused on strengthening the core infrastructure for reliable tool use, task automation, and model capability handling. I worked on bridging the gap between small local models and advanced agent workflows through both new feature development and critical bug resolution. ### Key Contributions (Local AI Focus): * **Capability Discovery & Probing (New Feature):** Designed and implemented the background probing system that automatically detects whether a local model (via Ollama or similar) supports native tool calling or requires a fallback mechanism. * **Standardized ReAct Fallback (Infrastructure):** Developed the ReAct execution path, enabling lightweight local models to perform complex tool calls by following structured thought processes. * **Reliable Stream Interleaving (Bug Fixes):** Resolved several long-standing issues with how local models handle interleaved thoughts and actions in response streams, specifically fixing parsing errors for indented and unanchored markers. * **Persistent Model Caching:** Created a thread-safe caching layer to store discovered model capabilities locally, significantly reducing latency and redundant API calls across agent sessions. --- ## BunkerAI: Autonomous Offline Intelligence Canonical URL: https://karpizin.com/projects/bunker-ai ID: bunker-ai Period: 2026-02-... Summary: Autonomous AI-server with a full human knowledge base for critical infrastructure, isolated environments, and space missions. Tags: Offline AI, Air-gap, Critical Infrastructure, Cybersecurity, Knowledge Base, Bunker AI Technology: Python, FastAPI, Llama.cpp, Whisper (Offline), Kokoro TTS, PostgreSQL, pgvector, RAG, Docker Website: https://bunkerai.ru External links: - BunkerAI.ru: https://bunkerai.ru BunkerAI is a high-reliability autonomous AI solution designed for air-gapped environments where external connectivity is impossible or strictly prohibited. It serves as the 'last line of defense' for knowledge preservation and decision support in critical infrastructure, deep-sea facilities, and space missions. ### Key Capabilities: * **Offline LLM Intelligence:** Integrates large language models (7B to 70B parameters) like Qwen, DeepSeek, and Mistral, running entirely on local hardware with zero external dependencies. * **Universal Knowledge Repository:** Contains a full mirror of Wikipedia (130GB+) and a RAG (Retrieval-Augmented Generation) system for private technical documentation and regulatory protocols. * **Voice-First Interface:** Features offline speech-to-text (Whisper) and text-to-speech (Kokoro) for hands-free operation in industrial environments, high-noise zones, or when wearing protective gear. * **Isolations & Security:** Engineered for Air-gap architecture, ensuring no data ever leaves the secure perimeter. It can integrate directly with local SCADA or ERP systems to monitor anomalies in real-time without cloud access. --- ## DataClerk.cc: Intelligent Document Ecosystem Canonical URL: https://karpizin.com/projects/dataclerk ID: dataclerk Period: 2026-01-... Summary: An autonomous AI-driven document management and analysis system designed for SMEs, NGOs, and educational institutions. Features advanced Agentic RAG, multi-LLM synthesis, and high format versatility. Tags: AI Agent, LangGraph, Agentic RAG, Automation, OCR, VLM, B2B SaaS, Privacy-First, Solopreneur Technology: Python, Vue 3, LangChain, LangGraph, LLM (GPT-4o, Claude 3.5), vLLM, Ollama, ChromaDB, Tesseract OCR, OpenRouter Website: https://dataclerk.cc DataClerk.cc is a next-generation intelligent documentation system that evolved from an experimental AI Sommelier into a robust, general-purpose business intelligence tool. It solves the critical challenge of extracting value from fragmented, heterogeneous data sources. ### Core Capabilities - **Maximum Format Versatility**: Seamlessly handles PDF, DOC, DOCX, XLS, XLSX, RTF, TXT, MD, CSV, along with photographs and scans (JPG, PNG) via advanced OCR and VLM (Vision-Language Model) integration. - **Agentic Reasoning (LangGraph)**: Unlike linear RAG systems, DataClerk utilizes complex state-machine logic to perform multi-step research, reflection, and data verification. - **Collaborative Multi-LLM Synthesis**: Implements a 'consensus' architecture where queries are processed by multiple top-tier models simultaneously, with results synthesized into a single, high-confidence answer. - **Dynamic Reporting**: Generates both free-form summaries and structured documents (reports, contracts, aggregations) using professional templates. - **Deployment Flexibility**: Operates both as a high-performance Cloud service and as a fully local, privacy-first installation (via vLLM/Ollama) for security-sensitive organizations. ### Vision As a **solopreneur** project, DataClerk.cc reflects a vision of the future where AI isn't just a search tool, but an autonomous 'digital clerk' capable of understanding, organizing, and reporting on the entire knowledge base of an organization. --- ## BHDT Global: China Sourcing Operations Canonical URL: https://karpizin.com/projects/bhdt-global ID: bhdt-global Period: 2025.12-... Summary: A production B2B website for structured China sourcing: supplier research and verification, quality control, logistics coordination, and a human-reviewed AI tooling roadmap. Tags: B2B, China Sourcing, Supply Chain, Supplier Verification, Quality Control, International Trade, Product Strategy, Web Development Technology: HTML, CSS, JavaScript, Nginx, Docker, Traefik, Let's Encrypt, Responsive Design, SEO Website: https://bhdt-global.aigrid.cc/ External links: - BHDT Global Website: https://bhdt-global.aigrid.cc/ BHDT Global is an English-language B2B website for companies building a supply chain in China. It explains a documented route from a product brief and factory research to supplier verification, quality control, shipment, and post-delivery follow-through. ### From concept archive to production The local project archive contained four visual concepts, a prototype, three levels of technical specification, commercial models, analytics notes, brand guidance, and a broad AI feature roadmap. I consolidated these materials into one coherent public product, keeping the credible operating model while removing mockup-only claims about customer counts, response times, factory rankings, and guaranteed outcomes. ### Information architecture The homepage presents the sourcing workflow as a working project rather than a generic list of services. Separate pages explain capabilities, the six decision gates, and the AI tooling roadmap. A deliverables interface shows what the client receives at each stage: a normalized supplier shortlist, verification memo, inspection report, and shipment handover file. ### Human-reviewed AI direction The AI page clearly distinguishes the roadmap from currently available services. Smart RFQ building, capability-based supplier matching, landed-cost scenarios, document review, visual QC pre-checks, and bilingual negotiation support are framed as tools that surface assumptions and open questions while keeping people responsible for verification and decisions. ### Production delivery The site is a lightweight static build with responsive layouts, canonical metadata, Open Graph assets, robots.txt, sitemap.xml, a privacy notice, and an inquiry form connected to the shared lead workflow. It runs in an isolated Nginx container behind Traefik with automatic HTTPS and a health endpoint. ### Value The result turns a large collection of strategy and design work into a focused international B2B presence. Instead of selling through unsupported promises, BHDT Global explains the process, evidence, boundaries, and next action a buyer can evaluate. --- ## EVQuery.com: China EV Specs and Comparisons Canonical URL: https://karpizin.com/projects/evquery ID: evquery Period: 2025.10-... Summary: An English-language catalog and comparison tool for Chinese EVs, EREVs and plug-in hybrids, with China-first pricing, trim data, galleries and public error reporting. Tags: EV, China EV, Electric Vehicles, Catalog, Comparison Tool, Automotive Data, SEO, Dark Theme, Vehicle Database Technology: Python 3.11, FastAPI, Server-side HTML Rendering, Vanilla JavaScript, SQLite, PostgreSQL, Vehicle Catalog, OpenGraph, SEO, Analytics, Playwright Website: https://evquery.com External links: - EVQuery.com: https://evquery.com EVQuery.com is an English-language automotive data product for researching Chinese electric cars, EREVs, plug-in hybrids and related new-energy vehicles. The product is built around a fast searchable catalog, model pages, trim tables, real photo galleries, and a comparison workflow for buyers, analysts, writers and EV enthusiasts. ### Production Snapshot The live site currently exposes **79 brands, 558 models, and 1535 trims**. It focuses on China-market specifications and China-first pricing, with approximate USD values shown where the exchange-rate metadata is available. ### Catalog Experience Users can search by model or brand, filter by powertrain, body style, China price, range, production years and practical model attributes. The interface is designed for quick scanning: a dense table on the left, model details on the right, sticky navigation, dark/light theme support, and responsive layouts for desktop and mobile. ### Model Pages Each model page combines factual summary cards, trim-level data, exterior and interior galleries, buyer-oriented notes, feature highlights, source-aware fields, and similar-model suggestions. Public pages are optimized for indexing with canonical URLs, OpenGraph images, structured metadata, and readable fallback content. ### Comparison Workflow The comparison tool lets users shortlist 2-4 models and compare metrics side by side: price, body dimensions, powertrain, range, performance, trims and feature signals. Shareable comparison URLs preserve the selected model set. ### Quality Loop Visitors can report factual, pricing, trim, photo, translation or outdated-information issues directly from the public interface. This turns the catalog into a living data product rather than a static brochure. --- ## AI Legal Pipeline: Risk Investigation and Automatic Document Filling Canonical URL: https://karpizin.com/projects/ai-legal-pipeline ID: ai-legal-pipeline Period: 2025-... Summary: A legal automation pipeline for investigating document risks, extracting obligations, preparing review notes, and filling standard legal forms with traceable AI assistance. Tags: AI Legal, LegalTech, Risk Investigation, Document Automation, Automatic Document Filling, LLM, RAG Technology: Python, LLM, RAG, React, FastAPI, PostgreSQL, OpenRouter, Document Parsing, Workflow Automation AI Legal Pipeline is an applied LegalTech workflow for document intake, risk investigation, and automatic document filling. The system uses LLM and RAG components to inspect legal documents, surface risky clauses, extract structured facts and obligations, and prepare draft outputs for standard forms and review workflows. The project focuses on repeatable legal operations rather than a generic chat interface: uploaded documents are parsed into structured evidence, checked against known risk patterns, summarized for review, and connected to downstream templates where extracted facts can be reused for automatic form completion. Key capabilities: - Risk investigation across clauses, obligations, dates, parties, penalties, and missing terms. - Evidence-backed review notes with references to the original document text. - Structured extraction for counterparties, contract metadata, duties, deadlines, and financial terms. - Automatic document filling for standard legal forms and internal review templates. - Human-in-the-loop boundaries for approvals, escalation, and final legal review. The pipeline is designed for teams that need faster legal document triage, cleaner intake data, and reliable preparation of repetitive paperwork without losing auditability. --- ## AURA.Home Canonical URL: https://karpizin.com/projects/aurahome ID: aurahome Period: 2025-... Summary: A system that transforms an automated home into a truly smart home. Tags: smart home, AI, IoT, sensors Website: https://aurahome.tech External links: - Website: https://aurahome.tech - GitHub: https://github.com/vcb88/aura-home AURA.Home is a concept and prototype of a next-generation smart home platform that shifts from the 'remote control' paradigm to a 'caring butler' paradigm. Unlike traditional systems (Home Assistant, Google Home), where the user has to manually configure scenarios, AURA analyzes data from multiple IoT sensors (light, temperature, humidity, presence, CO2) and learns the habits of the residents. The system automatically adjusts the climate and lighting, anticipating the user's wishes, creating a maximally comfortable environment without the need to reach for a smartphone or give voice commands. --- ## EverSoul: Neural Heritage & Digital Immortality Canonical URL: https://karpizin.com/projects/eversoul ID: eversoul Period: 2025-... Summary: A premium AI service createing intelligent digital twins (gravatars). Preserve the voice, manner of speech, and wisdom of loved ones for future generations using advanced neural heritage technologies. Tags: AI, Digital Twins, Gravatar, Neural Heritage, Voice Cloning, Legacy, B2C, Deep Learning Technology: Astro, Vue.js, GSAP, Python, PyTorch, ElevenLabs, OpenAI, Cloudflare Website: https://eversoul.ru External links: - EverSoul Website: https://eversoul.ru EverSoul is a pioneering service at the intersection of AI and personal legacy. It focuses on creating **"Intelligent Digital Twins"** or **"Gravatars"** — AI personalities based on a person's digital archives, voice recordings, and writings. ### Core Features: * **Neural Personality Backup:** Using LLMs to capture and replicate a specific person's manner of speech, knowledge, and life philosophy based on provided data. * **Voice Synthesis:** High-fidelity voice cloning that allows the AI twin to speak exactly like the original person. * **Digital Immortality:** A secure, private platform where families can interact with the digital twin of their ancestors, preserving wisdom and connection across generations. * **Privacy-First Legacy:** Built with deep encryption and strict access controls to ensure the digital heritage remains within the family. The project addresses the emotional need for connection and the technological possibility of preserving a person's essence, moving beyond static photos and videos to interactive, thinking avatars. --- ## NextThingDone: Essentialist Productivity System Canonical URL: https://karpizin.com/projects/nextthingdone ID: nextthingdone Period: 2025-... Summary: Gamified productivity ecosystem aimed at cultivating healthy habits and reducing cognitive load. Tags: Coaching, Psychology, Time Management, GTD, Gamification, AI, Microservices, i18n Technology: Golang, Vue 3, TypeScript, Mobile, Android, iOS, Telegram, Pinia, Cloudflare, FastAPI, Python, RabbitMQ, PostgreSQL, Docker, Gemini 2.0 Flash, Web Speech API Website: https://nextthingdone.com External links: - Telegram Journal: https://t.me/nextthingdone - Twitter (X): https://x.com/nextthingdone NextThingDone is an innovative task management and personal productivity ecosystem designed with a focus on minimalism, deep psychological engagement, and gamification. The project aims to eliminate cognitive load and help the user gradually become a better version of themselves through the philosophy of Essentialism. ### Key Features and Mechanics: * **AI Smart Voice Input (Dual-Mode):** A hybrid input system combining real-time browser transcription (Web Speech API) with multimodal audio analysis (Gemini 2.0 Flash). Users can dictate complex tasks ('Buy milk tomorrow at 9 AM'), and the AI automatically extracts structured data (Date, Time, Type) with high precision, using audio intonation as the source of truth. * **Gamification 2.0 'Golden Ratio':** A deep achievement system based on Fibonacci numbers. Includes: * **Progression:** Levels from 'Pair' (2 tasks) to 'Enlightened' (377 tasks). * **Shadow Side:** Anti-achievements for harmful patterns (e.g., 'Night Eater' for 3 AM work, 'Sisyphus' for undoing tasks). * **Willpower:** Streak tracking for habits ('Iron Rod') and app usage. * **Task Type Triad & Timezones:** * **Habits:** Strict 'Safe Noon' logic ensures habits remain pinned to their calendar date regardless of the user's timezone travel. * **Actions:** One-time tasks with precise deadline tracking. * **Projects:** Nested structures with 'Architect' achievements for decomposition. * **Global Architecture & Edge Performance:** Multi-region deployment ready (RU/Global) with separate billing integrations (Stripe/T-Bank). The network layer is optimized with **Cloudflare**, providing low-latency Edge Caching, Brotli compression, and high availability worldwide. Fully localized (i18n) into 4 languages. ### Tech Stack: * **Backend:** High-performance API on Golang (Gorilla Mux, Sqlx) handling core logic and gamification engine. * **AI Microservice:** Python/FastAPI service communicating via RabbitMQ to process tasks and generate coaching insights. * **Frontend:** Vue 3 + TypeScript SPA with offline-ready architecture foundation and reactive UI updates. --- ## CTOR Learning Platform Canonical URL: https://karpizin.com/projects/ctorlearning ID: ctorlearning Period: 2024.12-... Summary: Deep reinforcement learning platform for training and evaluating AI agents in the CTOR toroidal strategy game. Tags: Deep Reinforcement Learning, Game AI, Self-Play, Tournament System, ELO, Python, PyTorch, FastAPI, Next.js, Redis, SQLite Technology: Python, PyTorch, FastAPI, Next.js, React, Redis, SQLite, MongoDB, Docker, Streamlit, Playwright, GitHub Actions External links: - GitHub: https://github.com/vcb88/ctorlearning CTOR Learning Platform is a high-performance research and engineering system for training game AI through deep reinforcement learning. It is built around CTOR (Circular TORus), a two-player strategy game on a 10x10 toroidal board where edges wrap around and tactical patterns behave differently from ordinary grid games. ### AI Training The project implements self-play training loops, experience replay, model checkpoints, auto-resume, baseline opponents, and multiple neural architectures. Later versions added phase-aware inputs, differential reward shaping, future vulnerability penalties, connectivity rewards, and model families such as ResNet-style networks and TViT-Dual. ### Evaluation And Tournaments A distributed tournament system evaluates trained models against historical checkpoints and algorithmic players such as Minimax, SmartMinimax, Defensive, and Random. It includes ELO calculations, match scheduling, worker queues, tournament metadata, replay support, and frontend dashboards for rankings, system telemetry, and infrastructure status. ### Infrastructure The platform combines Python game-engine code, PyTorch training, FastAPI services, Redis queues and events, MongoDB metadata, SQLite double-write migration, compressed NPZ game storage, Docker profiles, monitoring dashboards, and Next.js frontends. The system is designed to run local experiments, scale game workers, preserve training progress, and make model quality measurable rather than anecdotal. ### Engineering Value CTOR Learning is a compact laboratory for agentic AI engineering: a real rules engine, adversarial environment, model lifecycle, observability, distributed execution, storage migration, and UX surfaces for playing, testing, and comparing AI agents. --- ## LocalLogic: Sovereign AI Agency Canonical URL: https://karpizin.com/projects/locallogic ID: locallogic Period: 2024-... Summary: A boutique AI development agency specializing in 'Sovereign AI'—on-premise, autonomous, and private artificial intelligence solutions that ensure 100% data privacy without cloud API dependencies. Tags: AI, Agentic RAG, LLM, On-premise, Privacy-First, Hardware, Consulting, B2B Technology: Llama, Qwen, vLLM, Ollama, Ubuntu Server, CUDA, Python, Docker Website: https://locallogic.ru External links: - LocalLogic Website: https://locallogic.ru LocalLogic is a specialized AI development agency focusing on **"Sovereign AI"**—delivering on-premise, air-gapped, and highly secure artificial intelligence solutions for enterprise clients. The core philosophy is to move businesses away from vulnerable cloud-based APIs (like OpenAI) to local ecosystems where the client owns the data, the hardware, and the intelligence. ### Core Services: * **Custom AI Development:** Building bespoke LLM agents and advanced Agentic RAG (Retrieval-Augmented Generation) systems that integrate seamlessly with internal databases, CRMs, and 1C. * **Turnkey Hardware Solutions:** Supplying and configuring specialized GPU server clusters (NVIDIA RTX/A/H-series and Apple Silicon Mac Studios) optimized for local LLM inference. * **AI Consulting & Audit:** Assessing ROI for AI integration, auditing existing implementations for security risks and "hallucinations," and fine-tuning models (like Llama and Qwen) on proprietary business terminology. The agency caters to privacy-conscious enterprises, critical infrastructure facilities, and specific verticals like LegalTech, FinTech, and B2B Operations, ensuring zero subscription fees for API usage and absolute data sovereignty. --- ## XPI-Blocks: The ROS2 & Raspberry Pi Building Blocks Library Canonical URL: https://karpizin.com/projects/xpi-blocks ID: xpi-blocks Period: 2024-... Summary: Comprehensive, modular, and perfectly documented library of examples (blocks) for ROS2 and Raspberry Pi, lowering the barrier to entry for robotics and enabling intelligent behaviors through LLM/VLM integration. Tags: ROS2, Raspberry Pi, Robotics, Framework, Embedded, Simulation, Drones, Reverse Kinematics, Sensors, AI Technology: Python, C++, ROS2, gpiozero, I2C, SPI, MQTT, smbus2, pyserial, Docker, LLM, VLM Website: https://github.com/karpizin/xpi-blocks External links: - GitHub Repository: https://github.com/karpizin/xpi-blocks XPI-Blocks is a comprehensive, modular, and perfectly documented library of examples (blocks) for ROS2 and Raspberry Pi. Its mission is to lower the barrier to entry for robotics and enable intelligent behaviors through LLM/VLM integration. Each block includes a clean ROS2 Node (Python/C++), hardware interface, launch file, and documentation (wiring diagram, BOM, topic interfaces, verification command). The project uses Ubuntu Server 22.04 LTS for Raspberry Pi, ROS2 Humble Hawksbill, and supports flexible LLM/VLM backends (Gemini, OpenRouter, Ollama). It provides standardized Hardware Abstraction Layers (HAL) for GPIO, I2C, SPI, and UART, and includes Dockerfile support for rapid deployment. --- ## LLM Monitor (OpenRouter Watchdog) Canonical URL: https://karpizin.com/projects/llm-monitor ID: llm-monitor Period: 2026-01 Summary: A high-performance observability tool for monitoring LLM availability and latency. Automatically discovers new free models on OpenRouter and provides real-time alerts via Telegram. Tags: LLM, Observability, Monitoring, OpenRouter, Telegram Bot, FastAPI, Python, Docker, CI/CD, SQLite Technology: Python 3.11, FastAPI, SQLite, Docker & Docker Compose, GitHub Actions, python-telegram-bot External links: - GitHub Repository: https://github.com/karpizin/llm-monitor LLM Monitor is a comprehensive watchdog system designed to track the health of free Large Language Models on the OpenRouter platform. It addresses the volatility of free tier availability by providing instant notifications when models go down or return online. Key Components: - **Automated Discovery**: Scans the OpenRouter API hourly to find and categorize new free models (including VLMs). - **Real-time Health Checks**: Performs smoke tests every 10 minutes to measure TTFT (Time To First Token) and overall availability. - **Telegram Integration**: A feature-rich bot allowing users to subscribe to alerts, check current status via /status, and receive instant push notifications on state changes. - **Web Dashboard**: A FastAPI-powered interface providing a clean overview of monitored models and their recent performance metrics. - **Production-Ready CI/CD**: Fully automated pipeline with GitHub Actions that builds 'clean' Docker images (no baked-in secrets) and performs secure SSH deployments using dynamic environment variable injection. --- ## Yandex: Alice AI Canonical URL: https://karpizin.com/projects/yandex-alice-ai ID: yandex-alice-ai Period: 2025-2026 Summary: Agent platform for the Alice voice assistant. Tags: agentic AI, platform, Alice, infrastructure, highload, AI, Yandex Technology: python, C++, infrastructure, OpenAI, Qwen, Minimax, GPT-OSS, GLM Leading the development of Yandex's internal platform, allowing the creation of complex, multi-step scenarios (skills) for the Alice voice assistant. The platform provided tools for dialogue management, integration with external APIs, and user behavior analysis. --- ## Archivarius: Intelligent File Librarian Canonical URL: https://karpizin.com/projects/family-archive-cataloger ID: family-archive-cataloger Period: 2024-2026 Summary: AI-powered system for deep analysis, cataloging, and semantic search across massive file archives and external drives. Tags: AI, VLM, LLM, Semantic Search, Archive Analysis, Computer Vision, Metadata, Deduplication, Python Technology: Python, FastAPI, SQLite, ChromaDB, OpenCV, VLM (Qwen-VL), LLM, Docker, PyMuPDF, Rawpy Website: https://архивариус.рф External links: - Archivus Pro (RU): https://archivus.ru - Archivus International (EN): https://archiv.us - Archivarius AI: https://archivarius.ai Archivarius is an intelligent platform designed to transform chaotic file archives into structured, searchable knowledge bases. Using a hybrid approach of fast heuristics and multimodal AI (VLM/LLM), it 'sees' photos, 'watches' videos, and 'reads' documents to determine their value and context. ### Core Capabilities: * **Multimodal AI Analysis:** - **Vision:** Uses VLM (Vision Language Models) to generate detailed natural language descriptions of images. - **Video:** Automatically extracts keyframes and creates collages for AI-driven content classification (detecting drone footage, dashcams, etc.). - **Books & Docs:** Full support for DJVU, FB2, EPUB, and PDF, including metadata extraction and semantic content summaries. - **RAW Support:** Native processing of Nikon NEF files using embedded preview extraction. * **Intelligent Quality Control (QA):** - Automatic detection of out-of-focus (blurry) shots using Laplacian variance. - Exposure analysis to identify underexposed or overexposed frames. - 'Junk' detection to filter out lens cap shots, accidental pocket photos, or empty textures. * **Semantic Search & UI:** - **Natural Language Search:** A dedicated Web UI section for finding files by meaning (e.g., 'mountain landscapes' or 'utility bills') using vector embeddings (ChromaDB). - **Interactive Dashboard:** Real-time statistics on archive health, value density scores, and processing status. * **Advanced Deduplication:** Identifies exact duplicates across different drives and directories, providing a primary-version strategy to reclaim storage space efficiently. Archivarius acts as a digital archaeologist, uncovering forgotten value in terabytes of data while automating the tedious task of manual sorting. --- ## sommelier.wine Canonical URL: https://karpizin.com/projects/sommelier-wine ID: sommelier-wine Period: 2024-2026 Summary: AI sommelier using LLM and RAG architecture for wine selection. Tags: AI, LLM, RAG Technology: python, linux, react, AI, RAG, LLM An assistant application that helps users choose wine. It is based on a Large Language Model (LLM) augmented with RAG (Retrieval-Augmented Generation) architecture, which allows the model to use an up-to-date database of wines, ratings, and reviews to provide accurate and relevant recommendations. --- ## Minimalist Web Portfolio Canonical URL: https://karpizin.com/projects/web-portfolio ID: web-portfolio Period: 2004-2026 Summary: A high-performance, minimalist personal portfolio showcasing over 20 years of engineering projects. Tags: Portfolio, Tailwind CSS, UI/UX, CI/CD Technology: Next.js, React, TypeScript, Tailwind CSS, Docker, GitHub Actions, Markdown Website: https://karpizin.com This project is the very platform you are currently browsing. It was designed to serve as a fast, clean, and professional showcase for a diverse engineering career spanning more than two decades. ### Core Philosophy: - **Minimalism:** A strict black-and-white theme to ensure project content and imagery remain the focal point. - **Performance:** Built with Next.js for server-side rendering and optimized image delivery, ensuring near-instant load times. - **Maintainability:** All project data is stored in modular JSON files, allowing for easy updates and translations without modifying the core engine. ### Technical Highlights: - **Next.js & React:** Utilizing the latest App Router and Server Components for optimal routing and SEO. - **Tailwind CSS:** Fully responsive design with a custom minimalist color palette. - **Dockerized Workflow:** The entire application is containerized, ensuring consistent environments between local development and production. - **Automated CI/CD:** Integrated with GitHub Actions for automated building, testing, and deployment to the server upon push. This portfolio reflects a commitment to quality, speed, and clean code, qualities that have been consistent throughout my work from 2004 to the present day. --- ## Live Neurons Canonical URL: https://karpizin.com/projects/live-neurons ID: live-neurons Period: 2025.02-2025.06 Summary: Concept of neural networks based on live neurons, modeled using Izhikevich models. Tags: AI, concept, neuroscience, biotech Technology: AI, python, neural networks Research concept project on modeling the behavior of biological neural networks. Using mathematical models (specifically, the Izhikevich model), a simulation was created demonstrating the emergence of complex activity patterns in the network, similar to processes in a living brain. --- ## NDA (e-vtols) Canonical URL: https://karpizin.com/projects/nda-evtols ID: nda-evtols Period: 2025 Summary: Development of a test bench based on Wirenboard, creating visual tools for test process monitoring. Tags: drones, evtol, ROS2, IoT Technology: python, javascript, wirenboard, nodered, dronecan, mqtt External links: - Wirenboard: https://wirenboard.com - Node-RED: https://nodered.org Within an NDA project, I am developing a hardware-software complex for bench testing electric aircraft (e-VTOL) engines. Key tasks: - Real-time telemetry collection from the test bench (current, voltage, RPM, temperature). - Data visualization for operators via a web interface (Node-RED Dashboard). - Integrating industrial Wirenboard controllers with drone avionics via MQTT and DroneCAN protocols. - Automating testing scenarios and data logging for subsequent analysis. --- ## Yandex: Agent Messenger Canonical URL: https://karpizin.com/projects/yandex-agent-messenger ID: yandex-agent-messenger Period: 2025 Summary: Development of a messenger agent to provide end-users with an option to delegate their tasks on communication with external services Tags: messenger, backend, Alice, Yandex Technology: python, infrastructure, AI, LLM, agents, Qwen, telegram Development of a messenger agent to provide end-users with an option to delegate their tasks on communication with external services --- ## CTORGame - Multiplayer Online Game Canonical URL: https://karpizin.com/projects/ctorgame-online ID: ctorgame-online Period: 2024-2025 Summary: Original multiplayer strategy game with a toroidal board, real-time play, rich tactical variation, and replay system. Tags: Multiplayer, Online Game, Real-time, TypeScript, React, Node.js, Redis, Socket.IO, Toroidal Board, Game Development Technology: TypeScript, React, Node.js, Express, Redis, Socket.IO, Vite, Tailwind CSS, Docker, Nginx, GitHub Actions, MongoDB Website: https://torgame.space External links: - demo: https://torgame.space - GitHub: https://github.com/vcb88/aura-home CTORGame is a modern multiplayer online strategy game with an original ruleset. It has a distant family resemblance to Tic-Tac-Toe in its pattern-building goal, but the actual gameplay is far more variable: players act on a 10x10 toroidal board, make multiple decisions per turn, manage automatic piece replacement, and navigate a much richer tactical space. The project demonstrates a deep understanding of distributed systems, real-time data processing, and interactive user interfaces. Players can create rooms, join by code, and play online using advanced mechanics. Key features: - **Original game mechanics:** 10x10 toroidal board, two operations per turn, automatic piece replacement, complex score tracking, and many more viable tactical branches than a simple grid game. - **Real online gameplay:** Creating rooms, joining by code, real-time updates, turn and status notifications. - **Reliability and persistence:** Automatic game saving, ability to resume interrupted games, anonymity, state preservation after server restart. - **Game history and replays:** Full game replays with VCR-like controls, turn navigation, strategy analysis. - **Scalable architecture:** Client-server model on React/Node.js, distributed state with Redis, Socket.IO for real-time communication, Docker for containerization, and CI/CD with GitHub Actions. --- ## Our (Un)Alternative Electric Future: A Comprehensive Course on Electric Vehicles Canonical URL: https://karpizin.com/projects/ev-course-2019 ID: ev-course-2019 Period: 2019-2025 Summary: A comprehensive course on electric vehicles, analyzing the EV market as of 2019 with updates up to 2025. Covers history, design, pros and cons, infrastructure, the global market, motorsport, and the future of EVs. Tags: electric vehicles, EV, electric transport, education, Tesla Technology: electric vehicles, batteries, electric motors, charging stations, IoT External links: - Full Course Program (Google Docs): https://docs.google.com/document/d/10d18VMyHEN6lEQA4Uo1alswap_SVCp4UsrbWuuP043c/edit?tab=t.0#bookmark=id.i7rxgqe9l86 Our (Un)Alternative Electric Future - A Comprehensive Course on Electric Vehicles (September 2019). The course provides a comprehensive analysis of the electric vehicle market as of summer 2019. At the time of creation, the EV market was in a phase of intensive growth, with China clearly leading, infrastructure actively developing, and battery costs continuing to decline. Key course sections: • Electric diversity — from micro-mobility to spacecraft • History of electric vehicles — eight stages of development (1828-2019) • EV Design — key components • Advantages and disadvantages — full analysis • Myths and Facts — debunking common misconceptions • EVs in Motorsport — records and achievements • Infrastructure — charging, transmission, and power generation • Global Market — production, sales, government support • Chinese EVs — from chasing Tesla to global dominance • Russia in the EV world — current state and prospects • Tesla and startups — new players in the auto industry • A Bright Future — trends for the coming years --- ## Cellar Technologies - Winery eCommerce & DTC Platform Canonical URL: https://karpizin.com/projects/cellar-technologies ID: cellar-technologies Period: 2024 Summary: A comprehensive cloud platform for wineries, combining an online store, wine club management, and inventory. Tags: E-commerce, DTC, Wine Industry, SaaS, FinTech, Inventory Management, POS Technology: TypeScript, React, Node.js, PostgreSQL, Redis, AWS, Stripe, iOS, POS Integration Website: https://ecom.getcellar.co External links: - Platform Website: https://ecom.getcellar.co - Company Site: https://getcellar.co Cellar Technologies is a modern SaaS solution designed specifically for the wine industry. The platform helps wineries grow Direct-to-Consumer (DTC) sales, automate inventory management, and improve customer experience. Key modules: - **E-commerce:** A powerful online store builder supporting complex shipping rules and alcohol taxation. - **Wine Clubs:** A flexible subscription management system allowing the creation of customizable wine clubs, shipment management, and automated billing. - **Inventory:** A unified inventory tracking system for all sales channels (online, tasting room, warehouse), preventing stock discrepancies. - **POS & CRM:** Integration with POS hardware and a customer profile system for personalized service. Support and development involved working with the integrated payment solution - Stripe; the system was actively used not only online but also in tasting rooms, offline stores, and at wineries. --- ## Cibotica - Automated Salad Maker Canonical URL: https://karpizin.com/projects/cibotica-salad-maker ID: cibotica-salad-maker Period: 2022-2024 Summary: Robotic kiosk for automated salad preparation. Tags: robotics, AI, foodtech, automation Technology: python, linux, ROS2, C++, baremetal, FreeRTOS, C++, Raspberry PI, Electronics, Vue, UX, sensors Website: https://cibotica.com Video: https://www.youtube.com/watch?v=EaUrFaaEMco External links: - Cibotica Website: https://cibotica.com - Techcouver Article: https://techcouver.com/2024/05/13/cibotica-salad-making-robot-creative-destruction-lab-seattle/ Cibotica (Remy) is a fully automated salad and bowl assembly line using AI and robotics. The system is capable of producing up to 300 servings per hour, ensuring precise ingredient dosing and minimizing food waste. The modular design allows the robot to be integrated into existing restaurant kitchens (under-counter installation). My role included: - Developing high-level robot control logic (ROS2). - Integrating a computer vision system to detect ingredient presence and quality. - Developing the Human-Machine Interface (HMI) for operators. - Optimizing manipulator motion planning algorithms to achieve maximum performance. --- ## Telegram Bots Suite Canonical URL: https://karpizin.com/projects/telegram-bots-suite ID: telegram-bots-suite Period: 2020-2024 Summary: A collection of multifunctional Telegram bots: from payment systems and stores to AI assistants and service utilities. Tags: Telegram, Bot API, Python, FinTech, Payment gateway, AI Technology: Python, aiogram, PostgreSQL, Docker, Redis, Github Actions, CI/CD, Telegram WebApps A portfolio of solutions developed on the Telegram Bot API platform, demonstrating the flexibility of using the messenger for business tasks. Examples of implemented projects: - **CryptoPay Bot:** A gateway for accepting cryptocurrency payments, generating invoices, and automatic conversion. - **SupportAI Bot:** A first-line support bot for a SaaS service. Uses RAG and LLM to answer user questions based on a knowledge base, switching to an operator only for complex cases. - **EventMaster:** An event organization system: participant registration, QR ticket issuance, notification mailing, and interactive schedule. - **ShopBot:** A full-fledged e-commerce store inside Telegram (Mini Apps) with a catalog, cart, and integration with payment providers. --- ## NDA - Artificial skin Canonical URL: https://karpizin.com/projects/nda-artificial-skin ID: nda-artificial-skin Period: 2023 Summary: Project to create artificial skin with haptic feedback (under NDA). Tags: robotics, haptics, NDA Technology: materials, electronics, firmware, robotics, AI Under a non-disclosure agreement, I worked on creating a flexible material mimicking human skin capable of transmitting tactile sensations. My area of responsibility included developing the software interface for controlling and interpreting signals from the sensor matrix. --- ## Droni.ca: Aerial Services, Mapping and Inspection Canonical URL: https://karpizin.com/projects/dronica ID: dronica Period: 2021-2023 Summary: A Canadian aerial-services brand and website covering drone photography, mapping, photogrammetry, infrastructure inspection, and data-driven field work. Tags: Drone Services, Aerial Photography, Mapping, Photogrammetry, 3D Modeling, Infrastructure Inspection, Computer Vision, Robotics, Canada, Service Website Technology: Docker, Nginx, Tilda, ROS2, Drones, Python, Computer Vision, Photogrammetry Website: https://droni.ca External links: - droni.ca: https://droni.ca Droni.ca was created as a Canadian aerial-services brand and presentation website for commercial drone work. The project combines field expertise, engineering credibility, and a clear service catalogue for customers who need imagery or spatial data from locations that are expensive, slow, or unsafe to inspect manually. ### Service model The offer is organized around four connected directions: aerial photography and video, drone mapping, photogrammetry and 3D reconstruction, and infrastructure inspection. These services cover marketing and real-estate imagery as well as construction documentation, terrain and asset mapping, volumetric analysis, agricultural observation, roof inspection, and visual assessment of hard-to-reach infrastructure. ### Mapping and photogrammetry High-resolution imagery captured from multiple angles can be processed into orthophotos, maps, terrain models, and detailed 3D representations of sites and structures. The website explains this workflow in business terms: faster data collection, repeatable documentation, more informed planning, and a reduced need for risky manual access. ### Trust and positioning Because drone services depend on safety and operational discipline, the site gives the team and qualifications a prominent role. The presentation includes Canadian UAV pilot certification, the Advanced Drone Operations exam, FAA Part 107 training, FPV training, an engineering background, drone experience since 2014, and photogrammetry experience since 2019. ### Website and brand layer The visual identity uses a custom drone-and-location mark, layered landscape graphics, an orange and green palette, and a compact navigation structure around the company, services, team, and contact path. The site was built as a focused lead-generation surface rather than a generic technology showcase. ### Technical delivery The public website was produced with Tilda and supported by a Docker/Nginx delivery layer. The wider technical context included Python, computer vision, photogrammetry, ROS2, and drone operations, connecting the marketing surface to practical data-processing and robotics expertise. ### Value Droni.ca packages a technically complex service into an understandable customer proposition: what can be captured, how the resulting data is used, why drone collection is safer or faster, and which qualifications support delivery. --- ## Platform for digital assets automated trading Canonical URL: https://karpizin.com/projects/ai-trade ID: ai-trade Period: 2020-2023 Summary: High-performance platform for automated real-time digital asset trading, designed for low latency and high throughput. Tags: HFT, Real-time, Trading, Digital Assets, FinTech, Highload, Blockchain, Web3, Crypto Technology: Go, Python, Kubernetes, Kafka, PostgreSQL, Microservices, Low Latency, Redis, Docker, Solana, Solidity This project is a comprehensive HFT (High-Frequency Trading) platform designed for fully automated trading in digital asset markets. The system is built on a microservices architecture using Golang for latency-critical components (matching, order execution) and Python for analytical modules and big data processing. Key aspects include real-time data processing, low-level optimization to minimize latency, integration with various exchanges via API, and robust infrastructure for 24/7 uninterrupted operation. The platform also includes modules for backtesting, risk management, and reporting. At its peak, the system operated with 3 exchanges simultaneously and executed up to 1 million trades per day. --- ## WayRay - AR HUD Canonical URL: https://karpizin.com/projects/wayray-ar-hud ID: wayray-ar-hud Period: 2014-2022 Summary: Development of a holographic AR display for cars with autonomous driving functions. Tags: Selfdriving, AR, Navigation, Automotive Technology: Python, Linux, ROS2, Computer Vision, Kalman filter, nagigation, map routing, OpenStreetMaps, sensor calibration Website: https://wayray.com Video: https://www.youtube.com/watch?v=IoLt--BNUhA External links: - WayRay Website: https://wayray.com - TechCrunch: https://techcrunch.com/2018/09/18/wayray-raises-80m-at-a-500m-valuation-led-by-porsche-for-its-holographic-ar-display-tech/ A project to create the world's first holographic AR display for cars (Deep Reality Display). The system analyzes data from car sensors (speed, navigation, ADAS data) and projects navigation prompts, obstacle information, and augmented reality elements onto the windshield at different depths without obstructing the driver's view. My role involved developing key algorithms: - Sensor Fusion (combining data from cameras, lidars, and CAN bus). - SLAM and vehicle positioning for accurate overlay of AR content on the real world. - Calibration of the projection system. The product evolved from a prototype (Navion) to an embedded solution for OEM manufacturers. --- ## RoboCharger - EV automatic charger Canonical URL: https://karpizin.com/projects/robocharger-ev-charger ID: robocharger-ev-charger Period: 2020-2021 Summary: Autonomous floor robot for wireless electric vehicle charging. Tags: robotics, AI, computer vision Technology: python, linux, ROS2, Computer Vision, AI, Electronics, aruco, Machine Learning A prototype concept for an autonomous wireless EV charging robot. Instead of plugging a connector into the side charging port, the device behaved more like a robot vacuum: it used computer vision to navigate under a parked electric vehicle, align itself with a wireless power receiver mounted in the lower part of the car, and start charging without manual cable handling. The system was designed around private garages, apartment parking, and public parking lots where a low-profile mobile charging unit could leave its docking station, locate the vehicle, position itself under the receiver, and make everyday EV charging feel almost invisible. --- ## WayRay Holograktor (Teleoperation) Canonical URL: https://karpizin.com/projects/wayray-teleoperation ID: wayray-teleoperation Period: 2019-2021 Summary: Vehicle teleoperation system using telematics and predictive AI. Tags: automotive, telematics, FPV, Starlink, Selfdriving, AI, 5G, VR Technology: python, linux, ROS2, Computer Vision, lidar, WebRTC Website: https://holograktor.com External links: - Holograktor Site: https://holograktor.com - Forbes Article: https://www.forbes.com/sites/alistaircharlton/2021/11/30/wayray-holograktor-is-a-remote-controlled-concept-car-built-for-the-metaverse/ Development of a teleoperation platform for the WayRay Holograktor concept car. This is a three-seater electric vehicle conceived as 'Metaverse on Wheels' for next-generation ride-hailing services. My team developed the remote piloting subsystem (Teleoperation). The system allowed an operator to drive the car remotely over 5G networks and Starlink using a VR headset and a dedicated workstation (cockpit), receiving panoramic video from the car's cameras with minimal latency. Key features: - Adaptive video stream compression depending on connection quality. - Predictive algorithms to compensate for control latency. - Integration with vehicle safety systems (automatic braking upon signal loss). - When the car is piloted remotely, the steering wheel retracts into the dashboard, freeing up space. --- ## Polygon Pathfinding Library Canonical URL: https://karpizin.com/projects/polygon-pathfinding-lib ID: polygon-pathfinding-lib Period: 2014-2018 Summary: A high-performance C++/Python library for pathfinding in open polygonal worlds using visibility graphs and navigation meshes. Tags: Pathfinding, Geometry, Visibility Graph, A*, Computational Geometry, Robotics, Game Dev, Python, TypeScript, curves, Bezier Technology: TypeScript, React 19, Vite, Tailwind CSS, C++, Python, Algorithms, A*, Bezier Splines Website: https://github.com/vcb88/pathfinder-demo External links: - GitHub Repository: https://github.com/vcb88/pathfinder-demo Development of a specialized library for finding the shortest paths in continuous 2D spaces with polygonal obstacles. The core algorithm is based on building a Visibility Graph followed by an A* search. Features included support for non-zero robot size via polygon expansion (Minkowski sum approximation), path smoothing using splines, and optimization for real-time applications in robotics and strategy games. This work laid the foundation for modern autonomous navigation modules. --- ## WayRay Element Canonical URL: https://karpizin.com/projects/wayray-element ID: wayray-element Period: 2014-2018 Summary: Smart car tracker with driving analytics and gamification functions. Tags: IoT, automotive, telematics, AutoYoga, Gamification, highload, wearable Technology: IoT, Bluetooth, OBD-II, Big Data, Java, Mobile, Android, AOSP, Firmware, Kalman filter, PostgreSQL, MongoDB, RabbitMQ, Python External links: - Startupticker Article: https://www.startupticker.ch/en/news/wayray-releases-its-first-product-wayray-element - IoT Evolution: https://www.iotevolutionworld.com/smart-transport/articles/414143-wayray-element-smart-car-tracker-plugs-into-on.htm WayRay Element is a compact device that plugs into a car's diagnostic port (OBD-II). It collects data on driver behavior, fuel consumption, and vehicle status in real-time. A unique feature of the project was the built-in gamification system ('Autoyoga'), which trained users to drive more smoothly and efficiently in a playful way. Data was synchronized with a mobile app via Bluetooth, where detailed statistics and recommendations were generated. --- ## LevelMeter: Precision Surface Tool Canonical URL: https://karpizin.com/projects/levelmeter ID: levelmeter Period: 2013-2016 Summary: A professional-grade Android utility that transforms a smartphone into a high-precision digital level and inclinometer. Tags: Android, Mobile Utility, Sensors, Accelerometer, Digital Level, Java, Tools Technology: Java, Android SDK, SensorManager, Audio Feedback API LevelMeter is a functional mobile application designed for construction, renovation, and DIY tasks. By leveraging the device's internal accelerometer, the app provides accurate measurements for surface leveling and angle calculation. ### Key Functionalities: - **Bubble Level Simulation**: A classic visual representation of a spirit level for horizontal and vertical alignment. - **Angle Measurement**: Real-time numerical display of inclination angles relative to the ground. - **Perpendicularity Check**: Specialized mode for verifying if two surfaces or objects are perfectly square. - **Audio Assistance**: Integrated sound-based feedback that changes pitch or frequency as the device approaches a perfectly level state (0°). This allows users to work without needing to look constantly at the screen, which is essential in awkward physical positions. - **Calibration System**: Custom offset calibration to ensure accuracy across different device hardware and casing variations. --- ## RoboDigest: Robotics News Portal Canonical URL: https://karpizin.com/projects/robodigest ID: robodigest Period: 2012-2015 Summary: A personal news portal dedicated to the world of robotics, featuring the latest updates, research breakthroughs, and industry trends. Built on WordPress with custom automation features. Tags: Robotics, News Portal, WordPress, Automation, Content Management Technology: WordPress, PHP, MySQL, HTML/CSS, Automated Tagging Algorithms, Newsletter Integration RoboDigest was a personal initiative to create a comprehensive Russian-language resource for robotics enthusiasts. The portal aggregated and published news about the latest robots, research papers, and technological advancements in the field. Key features included: - **Automated Content Management**: Implemented algorithms to automatically assign relevant tags to articles and news posts, improving navigation and SEO without manual overhead. - **Community Engagement**: Integrated a newsletter system to keep subscribers updated with the weekly highlights. - **Full-Stack Implementation**: Designed, developed, and maintained the entire project single-handedly using the WordPress ecosystem, customizing themes and plugins to fit the specific needs of a tech-focused news outlet. --- ## RemontSmartov.ru - Smart Car Service Canonical URL: https://karpizin.com/projects/remontsmartov ID: remontsmartov Period: 2009-2015 Summary: Development and support of a website for a specialized car service for Smart cars. Tags: Web Development, SEO, Automotive, WordPress, Car Service, Smart Cars Technology: PHP, WordPress, MySQL, HTML/CSS, jQuery, SEO Tools A project to create and develop the online presence of a specialized car service dedicated to the repair and maintenance of Smart cars (Fortwo, Roadster). The website became a key tool for attracting customers in a narrow niche. My tasks included: - **Development:** Creating the site based on WordPress CMS, adapting the theme to the automotive topic, creating a catalog of spare parts and services. - **SEO and Content:** Deep analysis of the semantic core (Smart engine repair, clutch replacement, diagnostics), writing expert articles, which allowed the site to reach top positions in search engines in Moscow. - **Functionality:** Implementing an appointment booking form, maintenance calculator, and a 'Q&A' section for consulting Smart owners. --- ## MobTrade: Sales Force Automation System Canonical URL: https://karpizin.com/projects/mobtrade ID: mobtrade Period: 2012-2014 Summary: An enterprise-grade mobile solution for sales representatives and field agents. The complex hardware-software system featured robust offline-first functionality, allowing agents to perform full daily operations without internet access, synchronizing with the central 1C ERP system and trading terminals via a custom replication protocol. Tags: Android, Enterprise, SFA, Offline-First, Data Replication, Java, SQLite, 1C Integration, Retail Technology Technology: Java, Android SDK, SQLite, Custom Replication Protocol, 1C:Enterprise Integration MobTrade was designed to optimize the workflow of sales representatives. As a key developer of the mobile client, I focused on high-performance UI and reliable data management. One of the project's most significant achievements was the implementation of a custom data replication protocol. This allowed agents to work fully offline in areas with poor connectivity, processing orders and inventory throughout the day, and then performing a bulk synchronization upon returning to the office. I was responsible for refactoring a legacy codebase with numerous initial bugs, implementing complex nested interfaces for fast data entry, and ensuring seamless integration with trading terminals and the 1C:Enterprise backend. The system successfully handled large product catalogs and client databases on the limited hardware of that era. --- ## AccelSnake: Gyro-Controlled Game Canonical URL: https://karpizin.com/projects/accelsnake ID: accelsnake Period: 2011-2014 Summary: A classic Snake game for Android featuring a unique control system based on device gestures. Unlike traditional tilt controls, it used a custom algorithm to analyze gyroscope angular velocity impulses, allowing users to play comfortably in any physical position (lying down, sitting) without recalibrating the 'zero' horizon. Tags: Android, Game Development, Gyroscope, Accelerometer, Sensor Fusion, Java, Legacy, Algorithms, Mobile Game Technology: Java, Android SDK (API 10), SensorManager, Ant, Eclipse ADT External links: - GitHub Repository (Restored): https://github.com/karpizin/accelsnake AccelSnake started as an experiment to solve a common frustration with tilt-controlled mobile games: the requirement to hold the device perfectly horizontal. I developed a custom gesture recognition algorithm that analyzed the history of gyroscope data to detect 'sway' impulses (acceleration followed by braking). This allowed the game to detect the *intent* to turn regardless of the device's static orientation relative to gravity. Technically, the project is a snapshot of early Android development (Eclipse, Ant, no Fragments). In 2026, I performed a 'digital archaeology' operation, restoring the full commit history from a dormant Mercurial repository and preparing a modernization plan to migrate the codebase to Android Studio and Kotlin. --- ## Ms. Pac-Man vs Ghosts AI Canonical URL: https://karpizin.com/projects/pacman-ai ID: pacman-ai Period: 2012-2013 Summary: Development of a competitive AI controller for Ms. Pac-Man to compete in the international 'Ms. Pac-Man vs Ghosts League'. Focused on real-time tactical decision-making and survival under uncertainty. Tags: AI, Game AI, Java, Python, Pathfinding, Tactical Planning, Ms. Pac-Man, Competitive Programming Technology: Java, Python, A* Search, Influence Maps, State Machines Website: https://pmsvg.github.io/ External links: - Ms. Pac-Man vs Ghosts League (Archive): https://pmsvg.github.io/ The Ms. Pac-Man vs Ghosts League is a prestigious series of academic AI competitions and a standard framework for game AI research, regularly featured at major conferences like IEEE CIG (Computational Intelligence and Games) and WCCI. Originating from the CEC 2011 competition, it quickly became a benchmark for AI agents, with over 60 Ms. Pac-Man agents participating in the 2012 edition alone. ### The Challenge The competition utilizes a Java-based simulator that closely mimics the original arcade game physics but introduces non-deterministic elements (ghost behavior, pill spawning). The core objective is to develop an autonomous controller capable of navigating the maze, surviving against a team of four ghosts, and maximizing the score through strategic pill consumption and 'ghost hunting' combos (200, 400, 800, 1600 points). ### Technical Constraints & Complexity - **Real-Time Limits:** Agents must make decisions within a strict 40ms time window per game tick. - **Observability:** While initially granting full observability, later iterations introduced Partial Observability (PO), where the agent only sees within a line-of-sight radius, requiring memory and prediction models. - **Adversarial Environment:** The ghosts operate as a coordinated multi-agent system (or individual controllers), making the environment highly dynamic and hostile. ### My Implementation My strategy focused on a hybrid approach suitable for the Full Observability track: - **Dynamic Pathfinding**: Utilizing optimized A* search to find safe routes while predicting ghost trajectories. - **Risk Assessment (Influence Maps)**: Developing a real-time 'heat map' of the maze where ghost positions projected negative influence (danger zones) and pellets/power pills projected positive influence (opportunity zones). - **State-Based Tactics**: Implementing a sophisticated state machine to switch between 'Foraging' (efficient eating), 'Evasion' (survival), and 'Hunting' (aggressive pursuit of vulnerable ghosts) modes based on distance metrics and timer states. This project not only achieved top-tier rankings in the league but also served as a practical exploration of techniques widely cited in research papers, such as MCTS and influence fields. --- ## PanoramaOSM: Open World Viewer Canonical URL: https://karpizin.com/projects/panoramaosm ID: panoramaosm Period: 2012-2013 Summary: An Android application for exploring the world through geotagged panoramic photos on OpenStreetMap. It served as an open-source alternative to proprietary Street View solutions, aggregating data from the (now defunct) Panoramio service and displaying it on `osmdroid` maps. Tags: Android, OpenStreetMap, Mapping, GIS, Java, osmdroid, Panoramio, Legacy Technology: Java, Android SDK, osmdroid, OSM Bonus Pack, REST API External links: - GitHub Repository (Restored): https://github.com/karpizin/panoramaosm PanoramaOSM was part of a family of 'PhotoPanorama' applications designed to visualize geotagged content on mobile maps. While other versions relied on Yandex MapKit or Google Maps v1, this project focused on the open-source ecosystem using OpenStreetMap and the `osmdroid` library. The app handled clustering of thousands of photo markers, asynchronous image loading, and map caching. Although the primary data source (Panoramio) was shut down by Google in 2016, the project remains a valuable reference for implementing custom map overlays and working with legacy mapping SDKs. I have recently restored the source code and documented a path to switch the data source to Wikimedia Commons API. --- ## Russian AI Cup: CodeTanks Canonical URL: https://karpizin.com/projects/russian-ai-cup ID: russian-ai-cup Period: 2012 Summary: Development of an autonomous tank AI for the inaugural Mail.Ru Group artificial intelligence championship. My Java-based strategy ranked in the Top 100 globally. Tags: AI, Game AI, Java, Robocode, Algorithms, Russian AI Cup, CodeTanks, Competitive Programming Technology: Java, Differential Steering Algorithms, Tactical Positioning, Target Selection Logic Website: https://russianaicup.ru/ External links: - Official Tournament Website: https://russianaicup.ru/ - Tournament Rules & Archive: https://russianaicup.ru/p/rules Russian AI Cup 2012: CodeTanks was one of the first major AI programming championships in Russia, organized by Mail.Ru Group and Saratov State University. Participants were tasked with writing a control strategy for a tank (or a squad of tanks) operating on a 2D battlefield. ### Technical Challenge The core of the competition was precise movement and fire control. Unlike simple 'WASD' movement, CodeTanks utilized differential steering where the strategy had to set the power for left and right treads independently. This required implementing custom physics-based movement algorithms to execute smooth turns and complex maneuvers while simultaneously calculating ballistics for both standard and premium projectiles. ### My Implementation (Top 100) Developed in **Java**, my strategy focused on several key areas: - **Advanced Positioning**: Used an evaluation system to identify 'safe' vs 'dangerous' zones based on proximity to enemies, potential lines of fire, and avoiding collisions with teammates. - **Dynamic Targeting**: A multi-criteria selection algorithm that prioritized targets based on distance, angle, and current health, ensuring high-efficiency fire. - **Tactical Maneuvering**: Implemented defensive 'jiggling' and reverse-tread movement to evade incoming fire while maintaining optimal turret orientation. Out of thousands of participants and a rigorous multi-round selection process involving team-based combat (up to 3 tanks per player), my bot successfully reached the final stages, finishing in the **Top 100**. --- ## Google AI Challenge: Ants Canonical URL: https://karpizin.com/projects/google-ai-challenge-ants ID: google-ai-challenge-ants Period: 2011 Summary: A global AI programming contest where I developed a distributed swarm intelligence bot for a real-time strategy game. Ranked in the Top 100 globally out of 7,000+ participants. Tags: AI, Swarm Intelligence, Game AI, RTS, A*, Minimax, Influence Maps, Distributed Computing, Google AI Challenge, Python, Algorithms Technology: Python, C++, Distributed Systems, Algorithms, Game AI Website: https://en.wikipedia.org/wiki/AI_Challenge External links: - GitHub Repository: https://github.com/karpizin/ants-ai - Wikipedia: AI Challenge: https://en.wikipedia.org/wiki/AI_Challenge The Google AI Challenge: Ants was an international programming contest held in 2011. My participation focused on two main areas: developing advanced swarm intelligence algorithms and building a scalable infrastructure for distributed testing. Key Highlights: - **Algorithmic Implementation**: Developed a sophisticated bot using **Influence Maps** for strategic territory control, **Minimax** with local clustering for tactical combat, and **A*** for efficient resource collection. These algorithms allowed thousands of ants to operate as a cohesive collective under strict real-time constraints (500ms per turn). - **Distributed Infrastructure**: Long before the ubiquity of Docker, I built a custom **Master-Worker architecture** to run matches across multiple machines. This system enabled rapid iteration by automatically benchmarking new versions of the AI against previous iterations to ensure consistent improvement in ELO rating. - **Performance**: Ranked in the **Top 100** globally among 7,000+ competitors, demonstrating the effectiveness of the combined algorithmic and infrastructural approach. --- ## Google AI Challenge: Planet Wars Canonical URL: https://karpizin.com/projects/google-ai-challenge-planet-wars ID: google-ai-challenge-planet-wars Period: 2010-2011 Summary: An international artificial intelligence programming contest where participants wrote self-contained computer programs (AI bots) to play a real-time strategy game called Planet Wars. As a developer, I ranked 4th in the tournament and also contributed as a volunteer supporting a matchmaking arena instance. Tags: AI, programming contest, game AI, RTS, strategy game, bot programming, Google AI Challenge, Planet Wars, algorithms, computer science, volunteer Technology: Python, Java, Algorithms, Game AI, Matchmaking Website: https://en.wikipedia.org/wiki/AI_Challenge Video: https://www.youtube.com/watch?v=A0ATZoeWV48 External links: - Wikipedia: AI Challenge: https://en.wikipedia.org/wiki/AI_Challenge - Planet Wars RTS Game Software (Updated Version): https://github.com/SimonLucas/planet-wars-rts - Gábor Melis' Planet Wars Post-Mortem (Rank 1): http://greg.bot.nu/planet-wars-post-mortem.html - Daniel Hartmeier's Unofficial Server & Stats: http://benzedrine.ch/planetwars.html The Google AI Challenge: Planet Wars was an international artificial intelligence programming contest held in Fall 2010. Participants developed self-contained AI bots to play a simple yet challenging two-player real-time strategy game called Planet Wars. The bots competed against each other on a server, with rankings determined by a Trueskill algorithm. Submissions were typically source code in various languages (Python, Java, C++, Ruby, Perl, Scheme, Haskell, and C#), which were then run in a Docker environment. My involvement was twofold: as a developer, I actively participated in the tournament, ranking 4th among thousands of entrants with my custom-built AI bot. Simultaneously, I volunteered to support the contest by maintaining an auxiliary matchmaking arena instance, ensuring smooth operation and providing additional testing opportunities for participants. This experience provided deep insights into game AI development, algorithmic optimization, and real-time system management. --- ## Conway's Game of Life: Colony Simulation Canonical URL: https://karpizin.com/projects/game-of-life ID: game-of-life Period: 2008-2011 Summary: A high-performance C++ implementation of Conway's Game of Life, focused on simulating the emergence and evolution of complex cellular colonies. Tags: Algorithms, C++, Cellular Automata, Conway's Game of Life, Simulation, Complexity Theory, Emergent Behavior Technology: C++, STL, GDI+/Graphics Library, Algorithm Optimization External links: - Wikipedia: Conway's Game of Life: https://en.wikipedia.org/wiki/Conway%27s_Game_of_Life This project is a dedicated exploration of cellular automata through Conway's Game of Life. Developed in **C++**, the primary goal was to create a robust engine capable of simulating thousands of generations for vast colonies of cells, observing how simple local rules lead to complex global behaviors. ### Core Simulation Features: - **Algorithmic Modeling**: Precise implementation of the four classic rules (survival, overpopulation, underpopulation, and reproduction) to model the life cycle of digital organisms. - **Pattern Cataloging**: Support for various well-known structures, including 'Still Lifes' (block, beehive), 'Oscillators' (blinker, toad), and 'Spaceships' (gliders). - **Advanced Constructs**: Successful simulation of complex engines like the **Gosper Glider Gun** and **Puffer Trains**, which demonstrate the ability of the system to produce infinite growth. - **Infinite World Concepts**: Exploration of different grid topologies, including toroidal (wrapping) maps to simulate unbounded space. ### Significance The project served as a deep dive into algorithmic efficiency and the study of emergent complexity, where unpredictable and beautiful patterns arise from a purely deterministic system. --- ## Monsters and Treasures (Social RPG) Canonical URL: https://karpizin.com/projects/monsters-and-treasures ID: monsters-and-treasures Period: 2010 Summary: A social isometric RPG dungeon crawler. Players explore procedurally generated dungeons, fight monsters, collect loot, and compete with friends. Tags: GameDev, Flash, RPG, Isometric, Social Game, Dungeon Crawler Technology: Java, MySQL, Flash, ActionScript 3.0, Hibernate, SmartFoxServer Monsters and Treasures is a social isometric RPG that brings the classic dungeon crawler experience to the browser. ### Core Gameplay - **Dungeon Exploration**: Players navigate through procedurally generated dungeons filled with traps, puzzles, and hidden treasures. - **Turn-Based Combat**: Strategic battles against a variety of monsters, from skeletons to dragons. - **Loot & Equipment**: A deep inventory system allowing players to equip weapons, armor, and artifacts to enhance their stats. ### Social Features - **PvP Arenas**: Challenge friends to duels to see who has the strongest build. - **Co-op Raids**: Team up to take down massive bosses for exclusive rewards. - **Leaderboards**: Compete for the top spot in weekly dungeon clear times and wealth accumulation. The game utilized an isometric projection engine built in Flash (ActionScript 3.0) and a robust Java backend using SmartFoxServer for real-time multiplayer interactions. --- ## CosmoZoo - Space Zoo Simulator Canonical URL: https://karpizin.com/projects/cosmozoo-game ID: cosmozoo-game Period: 2009-2010 Summary: An exciting space zoo simulator. Build habitats on asteroids, collect exotic species, and maintain the balance of alien ecosystems. Tags: GameDev, Flash, Strategy, Space, Social Game Technology: Java, MySQL, Flash, ActionScript 3.0, Hibernate CosmoZoo is a colorful real-time strategy game where the player manages a space zoo located on drifting asteroids. The player's task is not just to collect rare alien animals, but also to recreate suitable conditions for them: from gravity and atmospheric composition to flora and diet. Features: - **Procedural generation:** Each playthrough is unique due to the random generation of creatures and planets. - **Deep ecosystem:** Species interact with each other (predator-prey, symbiosis), requiring the player to maintain a delicate balance. - **Economy:** Station resource management, trading with tourist liners, and scientific research. --- ## Funny Gazomor Canonical URL: https://karpizin.com/projects/gazomor ID: gazomor Period: 2009 Summary: A humorous browser-based game styled as a vintage newspaper named 'The Red Valve'. Tags: Browser Game, Humor, Web Design, Stylization, PHP, JavaScript Technology: PHP, HTML/CSS, JavaScript, GD Library (Image Generation) A satirical browser game developed in 2009, distinguished by its unique visual design. The interface was entirely stylized to look like an old, yellowed Soviet-era newspaper titled 'The Red Valve' (Красный Вентиль). Players interacted with the game world through text-based articles and controls embedded within the newspaper layout. The project was a creative experiment in web styling and thematic storytelling, popular within specific LiveJournal communities at the time. --- ## Alice's Birthday (Game) Canonical URL: https://karpizin.com/projects/alices-birthday-game ID: alices-birthday-game Period: 2008-2009 Summary: Adventure game with platformer elements based on the cult sci-fi universe of Kir Bulychev. Tags: GameDev, Adventure, Platformer, Sci-Fi, Kir Bulychev, 1C Technology: C++, Game Engine, Scripting, Win32, DirectX Video: https://www.youtube.com/watch?v=YZbhCJOCrHI External links: - Wikipedia (Movie): https://en.wikipedia.org/wiki/Alice%27s_Birthday A video game developed to accompany the release of the full-length animated film 'Alice's Birthday', based on the novella by Kir Bulychev. The player controls Alice Selezneva, a girl from the future, who embarks on a space expedition to save the planet Koleida from a mysterious plague. The gameplay combines classic point-and-click adventure mechanics (solving puzzles, interacting with characters) with dynamic arcade and platforming sequences. Players explore futuristic Moscow, alien landscapes, and ancient ruins, using various gadgets to overcome obstacles. The project involved creating a faithful adaptation of the movie's art style and narrative within an interactive medium. --- ## Heroes of Might and Magic V Canonical URL: https://karpizin.com/projects/heroes5 ID: heroes5 Period: 2005-2008 Summary: Development of the iconic 3D turn-based strategy game Heroes of Might and Magic V, along with its major expansions. Tags: GameDev, AAA, Strategy, Heroes of Might and Magic, C++, Nival Interactive, Ubisoft Technology: C++, Game Engine, Build Engineering, Tools Development, Release Management External links: - Wikipedia: https://en.wikipedia.org/wiki/Heroes_of_Might_and_Magic_V - Steam: https://store.steampowered.com/app/15170/Heroes_of_Might__Magic_V/ Heroes of Might and Magic V is the fifth installment of the legendary turn-based strategy series, developed by Nival Interactive and published by Ubisoft. It was a landmark project, being the first in the series to transition into full 3D. I joined the project during its active development phase, initially contributing as a **Tools Programmer**, where I developed map editors and essential utilities for game designers. My role evolved as the project grew, leading me to become a **Team Lead** and eventually the **Release Manager** for the entire cycle. ### My Contributions: - **Lead Developer**: Managed the engineering and QA teams during the critical final stages of development. - **Build Engineering**: Standardized and automated the build process, ensuring consistent quality across multiple localized versions. - **Release Management**: Coordinated with Ubisoft to manage milestones, quality standards, and final delivery for the original game and both major expansions. - **Technical Localization**: Managed the technical complexity of localizing a content-heavy AAA game into dozens of languages. Working on the original game and its expansions, 'Hammers of Fate' and 'Tribes of the East', remains a career highlight, representing a successful revival of a beloved franchise. --- ## Paragon PenReader Canonical URL: https://karpizin.com/projects/paragon-penreader ID: paragon-penreader Period: 2004-2005 Summary: Cursive handwriting recognition technology for mobile devices (Windows Mobile, Android, iOS). Tags: Handwriting Recognition, OCR, Mobile, Windows Mobile, Symbian, Android, C++, Algorithms Technology: C++, Windows Mobile, Symbian, Android, Algorithms, Qt Website: https://www.paragon-software.com Video: https://www.youtube.com/watch?v=Fxqse6i_xGK External links: - Paragon Software: https://www.paragon-software.com Paragon PenReader is an advanced handwriting recognition technology designed for mobile devices. The project has a long history, starting from early PDAs (Apple Newton, Palm OS) to modern smartphones. Key features: - **Multilingual support:** Recognition support for over 40 languages, including complex scripts. - **Input modes:** Support for Cursive writing, Letter-by-Letter input, and Letter-on-Letter for small screens. - **Adaptability:** The system learns the user's handwriting features, improving accuracy over time. - **Cross-platform core:** The recognition core, written in C++, ensured high performance on the limited resources of mobile processors of that time (Windows Mobile, Symbian, early Android). --- ## SlovoEd Dictionary & Linguistic Engines Canonical URL: https://karpizin.com/projects/slovoed ID: slovoed Period: 2004-2005 Summary: Development of the industry-leading SlovoEd dictionary series for early mobile platforms (PalmOS, Symbian, Windows Mobile). Focused on advanced linguistic engines and high-efficiency data compression. Tags: Linguistics, C++, Pure C, Mobile, Compression Algorithms, PalmOS, Symbian, Windows Mobile, Slovomorpho, Slovosound, Linguistic Engines Technology: C++, Pure C, Algorithms, Speech compression During the mid-2000s, SlovoEd was the dominant dictionary software for the rapidly evolving PDA and smartphone market. My work focused on enhancing the core linguistic capabilities and resource efficiency of the software suite across PalmOS, Symbian, and Windows Mobile. ### Key Engineering Achievements: - **Slovomorpho Engine**: Developed a sophisticated morphology engine that provided grammatical support for dozens of languages. It enabled intelligent search, allowing the dictionary to recognize different word forms and provide accurate translations regardless of the conjugation or declension used. - **Slovosound Audio Base**: Designed and implemented the 'Slovosound' module, which integrated high-quality human-recorded pronunciations into the dictionary. This allowed users to hear correct pronunciations in various languages, a groundbreaking feature for mobile devices at the time. - **Extreme Data Compression**: Developed and optimized proprietary compression technologies to overcome the severe storage limitations of mobile hardware. Our innovation allowed a database of 50,000 words to occupy less than 20MB of storage—a record-breaking achievement that set the industry standard for performance and efficiency. - **Cross-Platform Architecture**: Maintained a unified codebase in C and C++ that ensured consistent performance and feature parity across diverse mobile operating systems and hardware architectures. --- ## ClubRadio (MIPT) Canonical URL: https://karpizin.com/projects/club-radio ID: club-radio Period: 2002-2004 Summary: A student internet radio station at MIPT that streamed mostly electronic music across the campus network and let listeners request and rate tracks. Tags: Community, Electronic Music, Streaming, MIPT, Phystech, Song Requests, Voting, Icecast, PHP Technology: Icecast, Shoutcast, PHP, HTML, Winamp DSP ClubRadio was a student internet radio project in the MIPT campus network in 2002-2004. Its address was **clubradio.7ka.mipt.ru**, and the stream focused mainly on electronic music. ### What listeners could do - listen to the music stream over the local network; - request a track through a web form; - vote for music; - follow track ratings; - see what had played recently. ### How it worked The service ran physically on personal computers in the MIPT campus. The web interface and the streaming setup connected early web technology with Shoutcast/Icecast and Winamp DSP. At the time, this made the campus network not only infrastructure, but also a shared musical space shaped by its listeners. The screenshots show a modern reconstruction of the ClubRadio interface in the visual language of the early 2000s, based on the surviving project description and remembered mint-green colour scheme. # Articles and external writing ## Why a Global EV Advisor Needs a Living EV Database URL: https://medium.com/@viacheslavkarpizin/why-a-global-ev-advisor-needs-a-living-ev-database-3b2a6aa7c017 Source: Medium Date: 2026-07-01 Summary: Why global EV advisory products need a living, source-aware vehicle database instead of static model lists: model identity, trims, markets, pricing, specs, and source confidence must evolve together. ## The Tyranny of Choice: Why Your 50-Item List is a Prison URL: https://medium.com/@viacheslav.karpizin/the-tyranny-of-choice-why-your-50-item-list-is-a-prison-a2c160c20423 Source: Medium Date: 2026-02-28 Summary: In 2000, psychologists Sheena Iyengar and Mark Lepper conducted the famous "Jam Study," demonstrating how having too many options can lead to decision paralysis rather than satisfaction. ## The Empathy Gap: Why You Treat Your Future Self Like a Rental Car URL: https://medium.com/@viacheslav.karpizin/the-empathy-gap-why-you-treat-your-future-self-like-a-rental-car-2b80957a011e Source: Medium Date: 2026-02-27 Summary: This article explores why we make poor decisions for our "future selves" due to a psychological disconnect that makes our future self feel like a stranger. ## The Architecture of Anticipation: Why Predictability is the Ultimate Productivity Hack URL: https://medium.com/@viacheslav.karpizin/the-architecture-of-anticipation-why-predictability-is-the-ultimate-productivity-hack-eea0bf48b1bd Source: Medium Date: 2026-02-25 Summary: Focusing on the "Anxiety of the Tomorrow Surprise," this piece discusses how even high performers can be terrified of their own task lists when they lack a clear view of what's coming next. ## The CPU of the Soul: Why Your Brain Needs a Garbage Collector, Not a Hard Drive URL: https://medium.com/@viacheslav.karpizin/the-cpu-of-the-soul-32964afcd628 Source: Medium Date: 2026-02-23 Summary: An exploration of why the human brain functions better as a high-performance processor (RAM) than a long-term storage archive, and the importance of mental garbage collection for productivity. ## The Electric Leap: How Chinese Semis are Outrunning the Future on China's Highways URL: https://medium.com/@viacheslavkarpizin/the-electric-leap-how-chinese-semis-are-outrunning-the-future-on-chinas-highways-d88d8f8dbeb6 Source: Medium Date: 2026-01-13 Summary: An analysis of the technological breakthrough in the electric truck industry and how Chinese manufacturers are overtaking the competition. ## The Death of the Budget Car: How the West Lost the Plot and China Rewrote the Rulebook URL: https://medium.com/@viacheslavkarpizin/the-death-of-the-budget-car-how-the-west-lost-the-plot-and-china-rewrote-the-rulebook-3f9173c59042 Source: Medium Date: 2025-12-20 Summary: An exploration of the decline of affordable Western cars and how Chinese manufacturers are redefining the budget vehicle segment. ## China's Automotive Blitzkrieg: 12 Cars from Guangzhou URL: https://medium.com/@viacheslavkarpizin/chinas-automotive-blitzkrieg-the-12-cars-from-guangzhou-that-are-rewriting-the-rules-dbf7da4d2011 Source: Medium Date: 2025-11-25 Summary: An analysis of 12 groundbreaking vehicles from the Guangzhou Auto Show that are setting new standards in the automotive industry. ## The History of Car Autonomy URL: https://medium.com/@viacheslavkarpizin/the-history-of-the-car-autonomy-d361ecf74fb2 Source: Medium Date: 2025-10-01 Summary: A look back at the key milestones and breakthroughs in the development of autonomous vehicles, from early concepts to modern self-driving technology. ## Recycling Fee 2025: Consequences for the EV Market in Russia URL: https://vc.ru/transport/2263545-utilsborn-2025-posledstviya-dlya-rynka-elektromobiley-v-rossii Source: VC.ru Date: 2025-09-23 Summary: An analysis of the impact of the increased recycling fee on the electric vehicle market in Russia in 2025. ## A Practical Guide to Choosing LiDAR Sensors in 2024 URL: https://medium.com/@slava.karpizin/a-practical-guide-to-choosing-lidar-sensors-in-2024-from-theory-to-implementation-51635197b906 Source: Medium Date: 2024-10-15 Summary: From theory to implementation, a comprehensive guide on selecting the right LiDAR sensors for your robotics and autonomous vehicle projects. ## How to Properly Charge an Electric Car: 5 Main Rules URL: https://dzen.ru/a/aanAKqun_xnMAb-m Source: Dzen Date: 2023-10-24 Summary: Essential guidelines for electric vehicle (EV) owners to maximize battery life and charging efficiency, including the 20-80% rule and the impact of temperature. ## Electric Car Results 2019 URL: https://vc.ru/transport/101037-elektroitogi-2019-goda Source: VC.ru Date: 2020-01-14 Summary: Overview of key events and achievements in the world of electric transport in 2019.