Right now I'm a Senior AI Engineer for a US biotech company, building autonomous agents that automate scientific research: agent orchestration, custom MCP servers, RAG over scientific papers, and the evals that tell us whether a change made things better or worse.
Next to that I build AI agents and MCP servers for businesses. I build everything with Claude Code. I don't read every line the agent writes — I prove it works: unit, integration and end-to-end tests, strict gates on every commit, and logs and traces in production. My part is the spec, the architecture and the quality bar.
Building
- Tovayo — a free, open-source AI front desk that books real appointments for small businesses.
- Airlock — an open-source gate that makes an agent ask a person before anything risky.
- Vedaword MCP — a public MCP server over 110,000 Sanskrit verses that says “no answer” instead of making one up.
- This site: express courses on AI engineering, its own MCP server at fedorthinks.com/mcp, and Markdown versions of every page for AI assistants.
Learning
- The stateless MCP spec (2026-07-28) and its extensions — tasks, apps, skills — and OAuth for agents.
- The enterprise stack my agents have to live in: Azure, Entra ID and Okta, Snowflake, Databricks and Fabric.
Thinking about
- How to trust code you didn't read: which tests, gates and production signals are enough to merge an agent's work with confidence.
- Where an agent should stop and ask a person — and how to make that step cheap enough that nobody turns it off.
Available for
- AI agent and MCP server projects — from the first call to launch. See services.
- Senior AI engineer or forward-deployed engineer roles, remote, with full overlap with US hours.