First Hour

Scammed online? Call 1930 first, then let The First Hour turn your screenshots into a ready complaint. Tamil and voice support for people unsure with English or official forms.

Comment

Inspiration

When someone loses money to online fraud in India, the first hour decides whether the money can be traced and frozen. The national cyber fraud helpline, 1930, exists for exactly this. But in that hour, victims panic. Many don't know to call 1930 first, what evidence to save, what the operator will ask, or how to write a complaint for the national cybercrime portal.

This is hardest for first-time digital users, elderly people, and small shopkeepers who are more comfortable in Tamil than English. Most existing tools focus on receiving complaints. I wanted to build something that helps the victim get ready to report.

What it does

First Hour is a mobile-first web app with a calm, guided flow:

  1. Emergency screen: a large button that dials 1930 immediately.
  2. Triage: 4-5 taps in Tamil (fraud type, amount, time, bank or UPI app, whether an OTP or PIN was shared).
  3. Action checklist: a fixed, rule-based list of what to do right now.
  4. Evidence extraction: the user uploads screenshots, and a vision AI model reads the transaction ID, amount, date, UPI ID, and phone numbers. The user confirms or corrects every field.
  5. Complaint draft: AI turns the confirmed details into an English complaint for the portal, a Tamil summary, and a one-screen call card to read out to the 1930 operator.
  6. Follow-up: the user can save the acknowledgement number locally and get a reminder to complete the portal complaint.

How I built it

  • Frontend: mobile-first web app with large buttons, with all Tamil text kept in one editable file
  • Backend: a small API endpoint that keeps the AI API key hidden
  • AI: a vision-capable LLM for screenshot extraction and complaint drafting
  • Voice: the browser Web Speech API for Tamil voice input and output
  • Privacy: no server-side storage. Case data stays in the browser and can be deleted with one tap

Key design decision: safety-critical advice comes from fixed rules, not AI. The AI only extracts details and drafts text, so it can never invent emergency instructions. Users also review every AI-extracted field before it goes into a complaint.

Challenges I ran into

  • Keeping AI out of safety-critical advice while still making it useful
  • Getting reliable extraction from varied, low-quality UPI screenshots
  • Tamil voice support differing across browsers and devices
  • Designing for panicked, low-literacy users with as few taps as possible

What I learned

  • Responsible AI design includes deciding where AI should not be used
  • Human confirmation of AI output matters in sensitive situations
  • For stressed users, simplicity beats features

What's next

More Indian languages, fraud-specific flows (loan app harassment, digital arrest scams), and WhatsApp integration.

Safety and limitation

First Hour is not a legal or police authority. It always directs users to 1930 and the official portal, never asks for OTPs or PINs, and does not promise recovery of money.

Built With

Share this project:

Updates

Submission history