From breaking software to building it.

I spent 6+ years as a QA Engineer, where my job was to find the ways systems break before users did, writing test plans, automating regression suites, and pushing back on releases that didn't meet the bar. That instinct, assume it's broken until proven otherwise, is exactly what's missing from a lot of AI agent projects shipped today.

I'm now focused on agent engineering: building systems with LangGraph and LangChain, wiring tools together with MCP, and grounding outputs with RAG. What sets my work apart is that I evaluate agents the way I evaluated software: with metrics, not vibes. Bug and error detection rates, false positive rates, and consistency across runs.

Right now I'm building this at my current company. Always up for trading notes with people working on agentic systems, especially where evaluation is treated as a first-class concern.