LinkSleuth helps a person decide what to do with a suspicious message.

You paste an SMS or an email, or you upload a screenshot. Gemma 4, an open-weight AI model from Google, reads it and points to the exact phrases that look risky. Our own code checks any links without ever opening them. A fixed panel then shows where to report in India: cybercrime.gov.in and helpline 1930.

It gives you evidence and a next step. When it can’t tell, it says so.

The problem

Picture an SMS that lands at 11 pm: “Your KYC expires today. Update now or your account will be blocked,” followed by a link.

Banks really do send KYC reminders, so the message could be genuine. The person holding the phone has a few seconds, a worried head, and advice they’ve heard a hundred times: “don’t click suspicious links.” That advice only works if you can tell which links are suspicious.

The scale is large. Government data reported in the press showed more than 16 lakh cybercrime incidents filed on India’s national reporting portal between 1 January 2020 and 7 December 2022. That count covers only the cases people reported.

Many spam filters and scam checkers return a label such as “likely spam” and stop there. A label doesn’t teach anyone anything, and it’s hard to trust a verdict you can’t check.

Who it’s for

Everyday users. The person who gets a strange SMS and wonders if it’s real. Also the relative who gets forwarded messages and is asked, “Can you check this for me?”

People who read in Hindi or Hinglish. Scams are written in the language people actually use. A tool for India should handle that.

Helpers, as a later idea. College IT cells, bank support desks, and community helplines field “is this real?” questions all day. We haven’t tested LinkSleuth with any of them yet.