Gregory B.

I build AI systems that run in production, not demos. Most recent proof: I took a casting platform from an empty repository to live iOS and Android apps in 16 weeks, on my own. It now holds 10,000+ artist profiles and runs an AI agent that writes casting briefs and matches artists to them, with explainability the client can audit. How I actually work with AI Implement, review and deploy run as separate agent identities across a nine-repository workspace. Each is its own unix account holding its own least-privilege token, so the separation of duties is enforced by the forge and the filesystem rather than by a rule an agent can forget: the lane that writes the code cannot merge it, the lane that deploys cannot land a pull request, and an approval only ever comes from a human hand. Lanes run headless on timers with fan-out ceilings and stale-claim sweeps. If you want agents doing real engineering work without handing them your production keys, this is the shape of it. What I do - AI agents and tooling: Claude and Claude Code, agent tool design, MCP servers, evals, prompt design, guardrails - Retrieval that actually ranks: hybrid search, pgvector embeddings, RAG on PostgreSQL, Meilisearch - Backends: Go (Gin), Python (FastAPI, Pydantic AI), OpenAPI-first APIs, PostgreSQL 17, Redis, S3 - Mobile and web: Flutter (App Store and Google Play), SvelteKit, Astro, Next.js, React - Infrastructure I run myself: NixOS, Kubernetes (k3s), Flux GitOps, OpenTofu, Cloudflare, Prometheus and Grafana Why clients hire me - One engineer, whole stack. No handoffs between a mobile shop, a backend shop and a devops shop. - I ship in weeks. The first working slice is usually live inside a month. - I know where LLMs earn their cost and where they quietly burn it. - I handle the non-code parts too: App Store review, GDPR, payment and legal flows. - Over 99.9% uptime on the systems I operate. Background 6 years shipping software with founders and end users. Contract work for a FAANG company, a US identity and cryptosecurity startup, and Nord Security (Go CLI and Flutter desktop client for NordVPN on Linux). CTO of a recruitment platform. Founder of wheretopark.app, an AI-over-CCTV parking system that took me to Regeneron ISEF in Los Angeles and won gold at IWIS Warsaw. Open source: ezsockets, a Rust WebSocket library (265 stars, in production use by third parties); houseflow, a home automation platform in Rust (190 stars); patches to tokio-rs/axum and helix-editor/helix. Polish (native), English (C1). Based in Gdansk, Poland. Comfortable with EU and US hours. If you have an AI feature that has to survive real users, real data and a real bill, that is exactly the work I do.