Hi everyone, I'm Diya, I built Glowby. I got into this from campaigning and interning at a state assemblymember's district office. Between doors and constituent calls, I kept meeting people who were certain about things that weren't true, and a lot of it traced back to a video they'd seen. So I tried to build a fact-checker. My first prototype was confidently wrong about everything. I asked it about peptides and muscle growth and it came back near 100% certain, citing a Stanford study about trees. The study did not exist. That's when I stopped tuning the model and started writing evidence rules instead. Glowby now splits the work across separate agents for vision, audio, claim extraction, evidence, and judging. Sources are ranked across 13 reliability categories. The headline score is the minimum across the central claims rather than the average, because averaging lets one big false claim hide inside a pile of true ones. Two weeks after launch TikTok and YouTube started blocking my servers, so there's now a fallback ladder of retries, proxies, and a paid scraper tier running underneath. I'd love feedback on the scoring rubric specifically. Methodology is public at glowby.io. Happy to answer anything.

Glowby

Paste a video link, get a 0 to 9.9 accuracy score

Paste a video link, get a 0 to 9.9 accuracy score