Generating college basketball matchup analyses by synthesizing Reddit fan commentary from team-specific subreddits, creating structured breakdowns grounded in fan perceptions.
A system that transforms unstructured Reddit discussions into coherent, position-by-position matchup narratives for college basketball games.
Automated scraping pipeline that collects fan commentary from Reddit's CollegeBasketball subreddit.
The scraper parses game thread index posts to identify relevant discussions, then recursively collects comments, user flairs, and metadata. It handles rate limiting and stores data in structured formats for downstream processing.
Relational schema optimized for querying fan discussions by team, game, and user context.
Core tables include users (with flair affiliations), posts, comments (with threading), teams, and games. Foreign key relationships enable efficient joins for retrieving team-specific commentary while handling data conflicts through upsert logic.
Retrieval-augmented generation system that synthesizes fan insights into structured matchup reports.
Comments are embedded using GTR-T5-XL model and stored in ChromaDB. For matchup queries, relevant fan discussions are retrieved via semantic similarity, then fed to DeepSeek model with structured prompts to generate position-by-position analyses grounded in user sentiment.
Delivered qualitative scouting reports that capture fan-driven narratives, providing unique insights beyond traditional statistics.
Successfully synthesized complex fan discussions into actionable matchup breakdowns, highlighting key battles and concerns that shape pre-game expectations.