At Kindo, I led post-training model adaptation, evaluation, and production release of a Qwen-based LLM for cybersecurity applications. My recent work has focused on adapting and evaluating models for production use.
At WET DESIGN, I designed and shipped an internal AI application that helped employees and executives investigate company data, answer operational questions, and generate business outputs. I implemented multi-turn Claude orchestration, governed read-only SQL execution, and evaluation and observability for grounding, failures, latency, and cost.
At STEALTH AI STARTUP, I built and deployed AI applications including Cicci, an AI-powered matchmaking platform, and Marvus, an AI-powered event networking platform. I designed Cicci’s embedding-based matching algorithm and validated it against a purpose-built golden dataset.
At Jet Propulsion Laboratory (NASA), I built ML pipelines and evaluation infrastructure across applied research programs. Earlier, at Quiet Machines, I designed and implemented computer vision and ML systems for robotics, while at California Institute of Technology I worked on a video-editing imitation learning system.

