Automate finding arrest-related news in the US to upload the police bodycam footage to YouTube. Here's how it works:
- The application queries the Google News API looking for arrest-related articles.
- It then categorizes and scores each arrest by passing the data through multiple LLM prompts, outputing a structured (pydantic-defined) result and discarding invalid or low-interest arrests.
- For arrests considered interesting, the application scrapes the internet to attempt to find all required information that isn't included in the analyzed article (e.g., the article could mention the suspect's name but omit the case number. This step would attempt to find it in other web sources).
- It implements multiple deduplication strategies, since often more than one news outlet will share the same story.
- After scoring, these results are exported into a Google Sheet and evaluated by humans for a possible FOIA request resulting in a YouTube video.
I ran myself a YouTube channel on this topic using this same script. I then presented this app to the biggest channel on boodycam footage on YouTube (~500K USD MRR) to use this script with their existing workflows.