Backend and platform engineer working mostly with Node.js and TypeScript. I build production services and distributed systems, and have shipped products across GCP, Alibaba Cloud, and AWS. My work includes self-hosted infrastructure, compliance audits, and leading engineering teams from early design through production.

I work with NestJS, CQRS and event sourcing, PostgreSQL, MongoDB, Redis, and messaging systems including RabbitMQ, Kafka, and NATS. I also build with Go, React, and React Native. These days a lot of my time goes into applied ML in Python: training small models and shipping them as CPU-only inference services.

Small CPU model that turn coding-workflow text into strict JSON: an 11-class schema router (logs, issues, commits, PRs, agent tool calls, diagnostics), a tool-match guard for agent runtimes (CALL / CLARIFY / ABSTAIN with verbatim argument filling), and zero-shot triage primitives. Fine-tuned encoders served as int8 ONNX under 2 GB RAM at ~8-16 ms per call, with REST and MCP serving. Apache-2.0, weights on Hugging Face.

An Nx and pnpm monorepo starter with NestJS API, Next.js web and admin apps, and an Expo mobile app. The API includes CQRS, multi-tenant guards, and a transactional outbox with Kafka dead-letter handling and replay. Shared packages cover auth, encryption, queues, storage, policy, and observability. The local Docker Compose stack includes PostgreSQL, Redis, Kafka, MinIO, and MailHog.

A Python monorepo for learning and building with the Python ecosystem, managed with uv workspaces. It includes a FastAPI service, a Typer CLI, shared Python packages, and a Next.js frontend. PostgreSQL with pgvector and Redis support the API and web app; Alembic manages database migrations.