A free scam shield for Indian elders and their families. Paste the message, upload a screenshot, or answer a few questions. Get a plain answer: what it is, why it is suspicious, and what to do next.
Open JaanchLoDigital-arrest scams alone took an estimated Rs 1,935 crore from Indians in 2024, and victims are disproportionately elderly. Most scam checkers are English-first, cloud-dependent, and blind to the patterns that actually hit Indian phones. JaanchLo is built for exactly that gap: it knows the Indian scripts, reads Hindi and Hinglish, and runs entirely on your device.
Eight scam families, learned from Indian cyber-fraud advisories and reported cases.
"CBI officer" on a video call, an arrest warrant, demands to stay on the line and pay a deposit.
"Your parcel has drugs in it." A customs fine, a police transfer, pressure to pay now.
KYC expiry, PAN update, electricity cut tonight. A link that looks almost official.
"I sent you money by mistake, please return it." Collect requests and QR codes that take instead of give.
A celebrity deepfake promising guaranteed returns, WhatsApp trading groups, apps that never pay out.
Easy tasks with small payouts first, then a deposit to unlock your earnings. Lotteries you never entered.
An instant loan, then harassment: morphed photos, threats to your contacts, demands that never end.
"Mummy, my phone broke, this is my new number, send money urgently." Sometimes with a cloned voice.
Paste the message, upload a screenshot (read on your device, never uploaded), or answer the live-call checklist.
48 signals across five dimensions: authority, fear, payment, channel, content. Twelve hard rules catch the unmistakable patterns; guards protect real messages from real banks.
Safe, Be Careful, or Danger, with reasons in plain English or Hindi and the exact next step, including when to call 1930.
Evaluated against 3,172 labeled messages with a held-out golden set that shares nothing with the rest. Every number is reproduced by one command in the open repository.
The false-alarm rate is the number we guard hardest. A shield that cries wolf gets uninstalled, and then it protects no one.
Honest limits. JaanchLo is a second opinion, not a guarantee. It cannot see who is really calling, and new scam scripts appear every month. When in doubt: do not pay, hang up, call 1930.
The detector, the scoring rubric, the 3,172-row English, Hindi and Hinglish scam corpus, the tests, and the CI gate are all public. Nothing like this dataset existed openly before. Fork it, audit it, improve it.
Code AGPL-3.0 · Dataset CC-BY-4.0 · Model, when it ships, Apache-2.0