A film screening aggregator website for independent film theaters in NYC powered by LLM agents.
Right now it's able to collect data from more than 30 sites with all very funky html formats with no custom code for each site.
When I began I had around 20% errors/hallucinations, right now it's way lower at around 3% errors in extraction. It's been fun and gave me a lot of experience building LLM powered data pipelines.