Building production-facing layers around generative models.
Agent runtimes, evaluation, fine-tuning, and observability to make model behavior easier to debug and ship, alongside exploring and learning RL techniques, physical AI and voice models.Recent wins in patent search and conflict resolution and physical AI orchestration in robotics at AWS Loft SF and Seeed Studio with NVIDIA and Hugging Face.
I turn uncertain model behavior into systems that can be inspected, evaluated, and improved. This site collects the projects, experiments, and notes behind that work.
San Francisco Bay Area