CV
Pavel Khimyack
Engineering and architecture: Python, event sourcing, AI
I design and write systems where data, money and permissions live in one model: an event-sourced core, projections for reads, business rules in a separate layer. I am the author and main developer of Tribe, a constructor for self-governing communities.
A second line is intelligent systems: LLM agents in a working loop with a run journal and metrics, MCP integrations, automation of data and reporting. I mentor at the University of Artificial Intelligence.
Open to remote product teams, contract or full-time.
Experience
Author and lead developer
2026 — presentTribe — a constructor for self-governing communities
A platform for communities: co-living houses, cooperatives, guilds, learning programs. From the event schema and domain boundaries to the interface and deployment.
- Event-sourced core, CQRS, a single projection; 29 domain modules and a process engine.
- Money modules: treasury, deals, guarantees with deposits and guarantors.
- Agent loop: MCP server, run journal, metrics, retro reviews.
- Production and development contours kept apart; canary checks and a heartbeat instead of waiting for complaints.
Mentor
presentUniversity of Artificial Intelligence
Mentoring in ML and software: reviewing tasks and submitted work, keeping the student track on course.
- Reviewing submissions against the stage checklist, discussing decisions and typical mistakes.
- Mentoring hours reported from three data sources instead of manual collection.
Stack
- Backend
- Python, FastAPI, Strawberry GraphQL, PostgreSQL, asyncpg, event modelling, CQRS, Alembic
- Frontend
- Vue 3, TypeScript, Vite, an in-house UI kit
- Infrastructure
- Docker Compose, Traefik, Let's Encrypt, CI, observability and health checks
- AI
- LLM agents, MCP, run orchestration, quality and cost metrics, data analysis
- Data
- SQL, collecting and reconciling data from calendars, messengers and a learning system
Projects
- Tribe platform — Event-sourced core, 29 modules, a process engine.
- Agents in production — MCP, run journal, metrics and retro reviews.
Languages
- Russian — native
- English — working level: reading and writing stronger than speaking
Files
You can save this page as a PDF through the browser print dialog — print styles are set up.