CyberSafe Buddy: A Scam Message Checker Built on My Phone

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

CyberSafe Buddy is a small web tool where you paste a suspicious SMS or WhatsApp message and get a risk level (Low / Medium / High), the red flags found, and a simple Hinglish tip on what to do next. I built it for people like my family and friends, who often receive fake prize, KYC and job messages on WhatsApp. I tested it myself on typical scam messages.

Demo

Try it here: https://huggingface.co/spaces/Tamsec/cybersafe-buddy

Note: the first load takes 1-2 minutes while the open-weight model downloads into your browser. Wait for "AI model ready ✅".

Example 1: a scam message
Input: Congratulations! You won ₹50,000. Click bit.ly/xyz and enter OTP to claim prize
Result: 🚨 Risk: High
Red flags: prize claim, OTP request, suspicious link
Advice: Don't open the link, don't share the OTP, block and report the message.

Example 2: a normal message
Input: Kal college 10 baje hai.
Result: ✅ Risk: Low
Red flags: none found

Code

The whole app is one index.html file in the Space (Files tab).
Link: https://huggingface.co/spaces/Tamsec/cybersafe-buddy/tree/main

How I Built It

• Built entirely from my phone as a Hugging Face Static Space (Gradio Spaces now need a paid plan).

• A rules engine (OTP requests, prize claims, urgency, suspicious links, fake job offers) decides the risk level.

• An open-weight model, Google FLAN-T5 (small), runs in the browser via Transformers.js and gives a yes/no "second opinion".

• My first version asked the model for a written explanation. It produced nonsense, so I limited it to a short yes/no question and let the rules make the final decision.

• Limits: it understands English and Roman Hinglish, not Hindi/Urdu script, and it is a helper, not a guarantee.

Why Does Open Innovation Matter?

Because the model is open-weight and runs in the browser, there is no API key, no cost, and the message never leaves the user's phone. That matters for a privacy-sensitive tool, and I can swap in a better open model later.