I build software by directing AI agents like an engineering crew: I own the spec, the boundaries and the quality bar, and the agents do the typing.
- Autonomous AI agent workflows for designing and shipping resilient software.
- Local RAG, LLM-powered automation and homelab infrastructure.
- Spec first: domain models, boundaries, edge cases and contracts are defined before any code is generated.
- Verified by behavior: quality comes from automated tests (unit, integration, E2E), deterministic benchmarks and runtime validation, not from eyeballing boilerplate.
- YAGNI: no needless abstractions, minimal dependencies, surgical diffs.
- GameTrack β Video game library and discovery platform with Steam sync, IGDB integration, personal stats and PWA support. (Live)
- MailShift β Privacy-first junk mail and newsletter cleaner for Gmail and Proton Mail, powered by local LLMs (Ollama / LM Studio) with concurrent IMAP processing.
Open to talking about agent architectures, open-source systems and collaborations.



