This weekend I went to Stanford TreeHacks 2025 (the biggest collegiate hackathon in the US), and it turned into a full-on “firsts” kind of trip: my first 36-hour all-nighter, my first time building a robot basically from scratch at a hackathon, and my first hackathon W. I fly out, do the whole pre-hackathon arc (General Catalyst Venture Fellows lunch, Apple Park rooftop with some wild AR glasses), then we check in and set up in the Data Science building to lock in. Our project was OmNom—an end-to-end food delivery robot. The core idea was simple: you type a one-sentence natural language request like “get me food from late night,” and the robot moves around and uses a stylus arm (6-DOF type energy) to interact with stuff. Night one was heavy hardware: chassis, linear rails, slider assembly—hand-drill-over-a-trash-can vibes. Day two was pure chaos: sensor fusion, IMU issues, mapping/object detection boxes, web interface, then the final stretch where everything starts crashing at 5:30 a.m. right when we’re filming the demo. We submit Devpost basically on fumes, roll into judging on zero sleep, and somehow we win Most Creative Hack. I’m proud because it wasn’t easy—this was one of those builds where you’re constantly debugging, reworking, and praying the bot moves at all… and then it actually does.