I built a platform that monitors 30+ production AI models and evaluates LLMs. It detects when live data drifts from training data distributions or when model performance unexpectedly degrades, and helps teams drill down to find the cause. Its LLM benchmarks help AI engineers make informed decisions about the best models for their use case.

I built an AI assistant that analyzes surgical robot logs and answers engineers' questions at an expert level. It coordinates multiple AI agents that pull the logs and context they need from internal data sources.

I created the core algorithms for Ouroboros, a tool that segments nerves and vessels in 3D scans too large to load into memory. It extracts only a thin, straightened volume along a traced path, segments it, and maps the result back into the original scan.

Graph diffusion models generate molecules and networks for drug discovery, but their memory grows with the square of the number of nodes. I led development of two models that reduce this cost. Scale-MGD trains with near-linear memory, and SparserDiff makes an existing model's edge sampling much faster and lighter.

I led the development of AI tools that turn teachers' lessons into flashcards, diagrams, and quizzes aligned to Namibia's national curriculum. The tools run inside #TeamGraduate's existing e-learning system, next to the lessons teachers already edit.

I built PLASMA, a web app that enables users to search millions of precomputed interplanetary trajectories. Users can visualize each trajectory in 3D, as well as options for launch vehicles and planetary approach.