
XO-Screens is a full-stack hackathon submission that brings together a user-facing interface and a containerized Track 2 video captioning pipeline. The application provides an interactive experience while the backend agent reads video tasks, analyzes each clip, and generates captions in multiple requested styles including formal, sarcastic, humorous-tech, and humorous-non-tech. The system uses Gemini as its primary vision and text engine, with Fireworks API support as a fallback for resilience. Designed for contest deployment, it runs inside Docker, writes structured results to results.json, and is built to be easy to test locally while remaining robust under automated evaluation.
13 Jul 2026