At DeepCare, I led development of medical AI agent products, including an Agent Harness for four medical departments. I built an agent loop with FastAPI and SSE that achieved ~98% tool execution success and 100% final JSON parseability.
For DeepCare’s orthodontic clinical decision support system, I worked with orthodontic specialists on agent workflows, backend engineering, and core algorithms. Across 260 clinical cases, over 90% of plans required no changes to key treatment decisions or tooth positions; I also developed PointNet-based dental landmark regression with 0.488 mm mean error.

