At Audacious Code, I build LLM and real-time voice agents and prove them with evaluations before they ship. For HeyJobs’ Madita.ai, I built the real-time interview engine and an AI-versus-AI test system; a model version change reached production with zero customer impact.
I own the quality bar end to end, from versioned prompts and LLM-as-judge evaluations to regression gates, token cost attribution, and p95 latency budgets. Since 2012, my CTO roles have also brought me into scoping problems with founders and customers, as well as debugging them in production.

