Sach: AI Fact-Checks WhatsApp Forwards in India
Summary
Pixel’s announcement introduces Sach, a tool that checks forwarded messages in English, Hindi, Telugu, and 14 other Indian languages against official sources. It gives cited TRUE, FALSE, MISLEADING, or UNVERIFIED verdicts, flags common scam tactics, and refers uncertain claims to human fact-checkers. The project shares its testing and evaluation details alongside a live demo and open-source code.
DEV Community
Every family WhatsApp group in India has the same problem
Every family WhatsApp group in India has the same problem: a forward lands, it sounds official, and nobody in the group can tell whether it is true. This week I built Sach to fix exactly that. Paste a forwarded message — in English, Hindi, Telugu or 13 other Indian languages — and Sach checks it against official sources, then returns a verdict you can forward back: TRUE, FALSE, MISLEADING or UNVERIFIED, with citations. Three things I refused to compromise on: Sach never guesses. FALSE requires a direct official contradiction. TRUE requires a direct official confirmation. If the evidence is not there, Sach says UNVERIFIED — and offers to hand the forward to human fact-checkers at PIB Fact Check and BOOM. It never takes credit for other people's work. If PIB Fact Check already debunked a claim, Sach returns that verdict with PIB's name on it. Every number is earned in the open. 5,125 automated tests. A 53-case red-team suite — prompt injections in three languages included. A published evaluation over real debunked forwards, with the runs in the repo, including the ones that exposed bugs. It also flags the shape of a scam — lookalike bank links, fake UPI refunds, .apk downloads, pressure tactics — separately from the verdict, in plain words. Built solo, with no out-of-pocket spend (free tiers plus Google Cloud trial credit), for the Code for a Billion — Bharat Agentic-AI hackathon. Submitted today; winners are announced Dec 5. Live now: https://sachcheck.in Watch the film (56s): https://lnkd.in/ecPtxUHX Code (MIT): https://lnkd.in/eV8ZApM4 Submission: https://lnkd.in/eY8329Vc Try it on the next forward your family group sends you. Then forward the verdict back. #BuildInPublic #AI #FactChecking #India

Powered by Readybot.io