AI-native Software Engineer with 5+ years of experience shipping production web products: an AI-powered unified inbox, B2B retail and fiscal platforms, international e-commerce and pharmacy ERP. Builds LLM-powered features in React and TypeScript (AI summaries with source evidence, RAG search, AI-generated tasks) over real-time WebSocket APIs. Works agent-first with Claude Code, Codex and Cursor: spec-driven development (OpenSpec), agent skills and MCP integrations, backed by automated quality gates. Open-source contributor to MobX.
AI & LLM Engineering:LLM APIs (OpenAI, Anthropic, Google Gemini), local LLMs (Ollama, Llama 3), RAG search, AI summaries with citations, structured outputs, tool calling, AI usage quotas
Product: FlowHuber (flowhuber.com), an AI-powered unified inbox for WhatsApp, Telegram, Slack, Microsoft Teams, Gmail and Outlook with AI summaries, search and tasks.
Core engineer on FlowHuber: own large parts of the React/TypeScript web client; contribute to the NestJS backend (AI summary pipeline adapters on RabbitMQ, Telegram integration).
Built the AI product surface: Highlights (AI summaries with priority, reasoning and source evidence), RAG-powered Smart Search and AI-generated Tasks with feedback loops that tune future suggestions.
Moved RAG search and Highlights from REST to WebSockets (Socket.IO); built WhatsApp/Telegram QR login, Gmail, Teams and Slack OAuth flows and live source statuses.
Shipped Stripe checkout, App Store subscriptions and AI usage quotas; led the UI migration to Tailwind CSS and shadcn/ui; built React Native bridge features and the Astro landing page.
Set up quality gates: Vitest/MSW integration tests, Storybook tests with a blocking a11y gate, coverage thresholds and bundle-size budgets; migrated ESLint/Prettier to Biome.
AI-native workflow: authored AGENTS.md/CLAUDE.md agent guidelines, plan changes as OpenSpec specs, implement with Claude Code and Codex, and feed agents Jira, Confluence and GitLab context via MCP.
Product: Smartpos Trade, a B2B retail platform integrated with tax systems and online fiscal receipts.
Developed the admin panel and merchant cabinet in React, TypeScript, Effector and Ant Design on Java/Spring REST APIs.
Delivered modules end to end: receipt statistics, billing, requisitions, promotions and referral programs, inventory with barcode scanning and Excel export.
Implemented real-time updates over WebSockets (STOMP), analytics dashboards (Highcharts), PDF export and i18n.
Adopted AI-assisted development (GitHub Copilot, ChatGPT, Claude) for refactoring, test writing and code review.
AI layer for a bill-splitting app: receipt understanding with an LLM and agent access over MCP.
Telegram bot (grammY) that reads receipt photos with Google Gemini, extracts items and discounts into structured data, asks who paid and creates bills automatically.
Open-source MCP server with 7 tools (TypeScript, Zod, Vitest) that lets Claude Code, Codex and OpenCode agents create and split bills, update payments and look up friends via the app's REST API.