I build and deploy AI systems for high-stakes clinical and scientific environments. My focus is at the intersection of foundation models, high-performance computing, and beneficial deployments in life sciences.

Currently, I'm a Senior AI Engineer at Genentech training foundation models for digital pathology. Previously, I built the algorithm for the first FDA-cleared AI to run directly on a bedside medical device at UCSF, and spent a decade running a computational biology nonprofit.

• NeuroAnswer: An AI copilot for exploring the MICrONs Minnie65 connectome dataset. Uses an MCP (Model Context Protocol) server to let researchers query 200k+ neurons and 500 million synapses via natural language, rendering the results directly into a 3D Neuroglancer scene.
• useful-transformers (RK3566 NPU Port): A low-level C++ optimization enabling transformer inference (Whisper) on highly constrained edge silicon. Engineered a fundamental rewrite of the library's matrix logic and ARM NEON intrinsics, refactoring the fp16 processing pipeline from an 8-wide to a 4-wide architecture to resolve hardware-level memory alignment crashes on the Rockchip RK3566 NPU. This port allows near-realtime whisper speech to text transcription on chips that cost as little as $10.

• Bedside AI for Critical Care (UCSF & GE Healthcare): Designed and trained the deep learning architecture for the first FDA-cleared AI algorithm fully integrated into a mobile medical device (detecting collapsed lungs on bedside X-rays).
  • 📄 Read the Publication (PLOS Medicine)
  • 📰 Read the Press Release (GE Healthcare)
• synominous: An in-browser vector search engine and interactive UMAP viewer for clustering ~100k fruitfly brain synapses, utilizing PGVector and off-the-shelf vision transformers.
• quarantine@Home: A distributed computing architecture I built to crowd-source small-molecule docking against COVID-19 viral proteins using AutoDock and GPU-optimized Docker containers.
• rsna-mammography: Implementations of Attention-based Multiple Instance Learning (MIL) applied to digital mammography screening.

• Project Infinome / Research Blog: Research, writing, and history from my decade running Infinome, a self-funded computational biology nonprofit.

• QuarantineAtHome was an emergency project that was rapidly built during the first two months of the covid pandemic. I lead the development of a "Seti@Home" and "FoldingAtHome" inspired drug docking experiment to find compounds that would target the spike protein and the Sars-Cov2 main protease. This project eventually got me a job working at Vir biotech where I developed a more refined internal docking pipeline.
• Hungrybot was arguably the first ever released AI food recognition app, a year before NotHotdog, and unlike NotHotdog, HungryBot also provided genomics analysis! This is probably the most insanely ambitious thing I ever built.