So this is really just a kinda proof of concept, but I'd be pretty easy to turn it into what you want. It sounds like you've already got both basically a script and images to go with it?
Yeah, that's suppose to be a Ken Burns style zoom but I haven't figured out how to stop the shake the should probably just remove it. And I need to remove the "By Claude" too. That's a remnant from porting from my internal framework for the [By Claude)(https://www.youtube.com/@byclaude) Youtube, TikTok, etc channels.
HHS released a massive dataset of every Medicaid payment to every provider in the US: 227 million rows covering $1.09 trillion in spending across 617,000 billing providers. The data was released explicitly to crowdsource fraud detection.
The raw data is a 2.9 GB Parquet file. I built MedicaidSpending.org to make it searchable and browsable.
You can search by provider name or NPI, browse by state/city/specialty, and see individual provider pages with monthly spending trends, billing code breakdowns, and automated billing flags for statistical outliers.
Some of the patterns are striking. Brooklyn alone accounts for $31.8 billion in personal care services (code T1019) _ more than most states spend on all Medicaid combined. Some authorized officials control hundreds of billing entities. Early analysts scanning just 0.16% of providers flagged $90 billion in likely fraudulent payments.
Technical details:
- Go single binary, ~15 MB
- 3.3 GB SQLite database (read-only, pre-aggregated from the 227M rows using DuckDB)
- 900,000+ indexable pages generated from 13 templates
- No JavaScript framework _ server-rendered HTML, Chart.js for one chart per provider page
- Runs on a single VPS behind Caddy
You sound super interesting and I’m sure plenty of people would be interested in chatting (including me!), so please think about putting your email in your profile!