Projects
F1TENTH: Automatic Emergency Braking
Brought up a 1/10-scale autonomous car and implemented automatic emergency braking (AEB) in C++ using time-to-collision (TTC). For every forward LiDAR beam the node projects the car's speed onto the beam, computes TTC = range / closing speed, and publishes a zero-speed Ackermann command the moment any beam falls under the braking threshold.
How it works GitHubScribe32: AI Clinical Scribe
An AI scribe that turns a recorded patient visit into a structured clinical note. In production with 11 daily users (one orthodontist and ten clinical assistants). It replaced a commercial scribe billed at $75 per user per month and runs for under $6 a month in total, about $0.45 per user.
Screenshots, stack and decisionsDental Pathology Segmentation
Built a Python and OpenCV pipeline converting COCO-annotated panoramic X-rays into nnU-Net training masks across 705 radiographs, then trained a nine-stage 2D U-Net on an NVIDIA H100, reaching a Dice score of 0.77 on impacted teeth (0.53 periapical lesions, 0.46 caries). Error analysis pointed to metallic implant artifacts as the main failure mode.
GitHub