What I built
- Rebuilt the scraper to read each venue's own website instead of one listings site. That took us from 275 gigs to almost 4,000.
- Made it scrape in parallel, with a polite mode that checks each site's robots.txt and waits between sites.
- Read gig flyers with OCR, in Japanese and English, from images and PDFs.
- Matched band names to Spotify artists so gig pages can play a preview, with GPT-4 helping pull the names out of messy listings.
- Also built the search by place, distance and genre, and the map view.
