Teaching machines to see, think, and move through the world.
Also hiking trails and baking things that may or may not turn out.
I'm a robotics and computer vision engineer who gets unreasonably excited when a system finally just works. The camera sees what it's supposed to see. The robot arm moves like it has intention. The math behind the chaos suddenly makes sense. I've spent the last few years building perception systems for medical devices, autonomous vehicles, aerospace, and sports analytics. When I'm not at a keyboard, I'm probably reading, trying a new bake, or thinking about how technology could be more useful for the people who actually need it most. ✦
I kept wondering what it would take for an autonomous vehicle to not just detect things, but actually anticipate where they're going. So I built a full pipeline: LiDAR point clouds, DBSCAN clustering, extended Kalman filters, all running in real time in CARLA simulation. Then I stress-tested it against published IEEE benchmarks, partly because trusting your own system blindly felt intellectually dishonest. Turns out the published numbers weren't always holding up either.
This one started with a simple question: what if a robot could help someone eat a meal on their own? A WidowX 250S arm, an Intel RealSense camera, MediaPipe watching for a mouth to open, and a ROS2 + MoveIt pipeline putting it all together. It sounds like a tidy description. It was months of debugging, recalibrating, and testing with real people until it actually worked. 90% success rate. Under $5K. Probably the most meaningful thing I've built.
Built a real-time CV pipeline for analyzing athlete movement on live figure skating footage. RT-DETR for detection, ByteTrack through all the occlusion chaos of an ice rink, RTMPose for biomechanical keypoints, and a custom ResNet-34 to segment the ice surface itself. Coaches get it straight to their phones. It's one of those projects where the domain made every technical problem feel worth solving.
People with congestive heart failure often wake up in the night struggling to breathe, and sometimes the right pillow angle is enough to help. Pillavate monitors SpO2 and heart rate through pulse oximetry and quietly adjusts its elevation with a servo motor. It's an Arduino project. It's not glamorous. But it's the kind of thing I find myself most drawn to: small, grounded, and actually useful to someone.
I've been lucky to work on things where getting it wrong actually has consequences.