The pipeline: ElevenLabs transcribes → Gemini generates → Gemini reviews (checks claims against source material, flags gaps) → publishes. The review step is the interesting part, it actually catches when the AI hallucinates beyond what the source said.
We've got 68 articles across 15+ towns. Most are places where the paper recently died. The cost to serve a town is under $5/year, which changes the economics completely, you don't need ad revenue or subscribers to justify covering a place.
Not pretending this replaces real reporters. But for a town where the alternative is literally nothing, a structured article from a resident's firsthand account beats the Facebook post that disappears tomorrow.
Hackathon project by a couple students, rough around the edges. Happy to talk about the technical side or hear why this is doomed.
Here's Chesterton, Indiana — their 141-year-old paper closed last year: https://news.minir.ai/explore?town=chesterton
Main site: https://news.minir.ai