GSoC 2026: Inquiries regarding #14 AIPlayerInsight

[xyx-creator]

Hi,

I'm Miles (Yuxuan), a Master's student at UCSB and incoming PhD at Notre Dame. My research focuses on Vision-Language Models (VLMs), which aligns perfectly with the goal of turning raw video into LLM-queryable insights for AIPlayerInsight.

I've carefully reviewed the required tech stack, but I have three quick questions to help accurately scope my proposal:

1. Codebase Status: I couldn't locate an existing repository online. Is this a greenfield project where we are expected to build the end-to-end pipeline entirely from scratch?
2. Performance Baseline: The expected outcomes mention "10x faster analysis by reducing time from whistle to data availability." Could you clarify what specific baseline or existing workflow this 10x improvement is being compared against?
3. Datasets & Evaluation: Will there be a standardized dataset (e.g., specific game footage) and defined evaluation metrics/benchmarks provided to measure the accuracy of the tracking and LLM retrieval?

Looking forward to your guidance!

Best,
Miles