ScamSaathi: an offline scam shield in Hindi and English

Every day people in India lose money to a fake "PM-Kisan.apk", an "RTO challan" file, a "KYC update" link, a UPI collect request or a fake scholarship, usually on WhatsApp or SMS, in a regional language, while they are alone. ScamSaathi checks a message, link or app name on the phone itself, in Hindi or English, explains why, and can tell someone you trust.

Three ideas built together

• Share-to-check: share a message from WhatsApp or SMS and get plain-language reasons, with a Listen button for people who read slowly. A suspicious .apk shows what it asks for (read SMS and OTPs, read the screen) without being installed.
• Offline and private by default: detection runs on the phone from signed rule packs. The message is never uploaded; reporting is opt-in and sends only hashes.
• The Saathi loop: name someone you trust; when something looks dangerous the app drafts them a short note that never contains the message, a link or a number.

How it decides

• Rules are data: phrases, link features and weights live in a pack, with every explanation in English and Hindi. A Dart and a Python engine must pass the same golden corpus, so they cannot drift.
• A small character n-gram model notices wording that is merely similar to scams; it can warn but never alone says "dangerous".
• No single clue is enough: at least two rules must agree. Nothing is ever called "safe"; a quiet result says "no warning signs found".
• Signed updates fail closed: packs are signed with Ed25519, and the app and server reject missing, tampered or older packs.

Honest numbers

Early release (v0.2). On a fresh, made-up test set it flagged 27 of 36 scams (75%) and 1 of 36 normal messages (2.8%). It has not been tested on real WhatsApp traffic, so every result is a strong hint, never proof.

Institutions such as a school, farmer-producer organisation or cooperative bank get a dashboard of what is circulating in their district and can publish verified alerts, without ever seeing messages or people. If you lost money, call 1930 or report at cybercrime.gov.in.