Electrical engineering and CS minor at the University of Florida. I mostly work on robotics and things that run on real hardware.
A robot arm that learns to pick things up. I recorded 160 demonstrations by teleoperating it by hand, then trained two policies: ACT got 18 of 20 autonomous trials, SmolVLA 16 of 20. One workspace position was kept out of training entirely and both policies still worked there. The setup costs under $300.
Videos and write-up. Along the way I patched a streaming bug in LeRobot, the library it runs on.
Research at USF. I trained a reinforcement learning agent to decode LDPC codes and it lost badly to a simple greedy baseline. Working out why turned up a bug in my own benchmark: a third to a half of the bits weren't covered by any parity check, so nothing could have decoded them.
An iOS navigation app for blind and low-vision users. It records a route as AR waypoints, replays it with spoken directions, and calls out obstacles with distance: "chair, 1.9 meters, left." Runs at 15 fps on an iPhone 14 Pro. Third place, Congressional App Challenge 2024. Demo.
Captain and lead programmer. I wrote our odometry and path following, including the 105-point autonomous routine that won the team its first Florida state title in seven years.