7+ years of systems and engineering experience in regulated enterprise environments. I build reliable workflows around messy operational work: backend services, data flows, AI-assisted retrieval, and safe automation with explicit human control where an external side effect matters.
Python backend · Data workflows · AI integrations · Telegram automation · Docker
An agentic support-workflow copilot with bounded autonomy and deterministic human policy gates. Built on the foundation of two earlier prototypes — Support Operator Panel (operator UI & backend) and assist-craft-qna (knowledge retrieval & reranking) — it refactors the domain into a modern agentic orchestration architecture.
What it demonstrates: stateful LangGraph orchestration, multi-source evidence grounding (runbooks, incidents, telemetry), Pydantic v2 contract boundaries, dual-engine deterministic fallback, immutable audit trail, FastAPI/PostgreSQL, and React 19 reviewer workspace.
Status: active reference implementation with 100% offline reproducibility and comprehensive automated test coverage.
A self-hosted Python/MCP control plane for provisioning isolated Telegram bots. A user explicitly confirms creation in Telegram; a persistent worker handles the credential outside the agent and MCP context.
What it demonstrates: approval boundaries, secret isolation, durable worker lifecycle, reconciliation for ambiguous external effects, packaging, CI, and PyPI release engineering.
Status: public alpha. Published on PyPI and listed in the Official MCP Registry; external adoption is not yet established.
A collaborative event-management platform with web and Telegram entry points, background workflows, and a containerized deployment stack.
My contribution: Python/Django backend, Telegram and aiogram integrations, Celery/Redis workflows, and deployment work.
Status: collaborative project — this repository does not claim sole authorship or a currently operating production service.
ML Model Compass is an explainable decision-support app for selecting practical ML baselines from a task description or local CSV. It is not AutoML and does not promise the best model.
Python · FastAPI · LangGraph · Django · PostgreSQL · REST APIs · semantic retrieval and reranking · Celery/Redis · Docker · TypeScript/React · CI/CD
- Telegram: @la5er
- Portfolio: larin.work