Skip to content

MirrorDNA-Reflection-Protocol/chetana-site

Repository files navigation

Chetana

Automated scam detection for Indian consumers -- SMS, WhatsApp, UPI, deepfakes, voice clones -- across 12 Indian languages.

Live License: MIT


Chetana is a check-before-you-act tool that turns suspicious digital content into a clear next step. Paste a message, upload a screenshot, or submit a UPI ID -- Chetana returns a verdict, an explanation, and a recommended action. It does not guess when evidence is weak: unclear inputs receive an unclear verdict, not a false confidence score.

Live at chetana.activemirror.ai.

How It Works

+------------------+       +------------------+       +------------------+
|   User Input     |       |  Verdict Engine  |       |     Response     |
|                  | ----> |                  | ----> |                  |
|  Text, image,    |       |  Deterministic   |       |  Verdict + type  |
|  QR code, UPI ID |       |  classification  |       |  + reasons +     |
|                  |       |  + evidence log  |       |  next action     |
+------------------+       +------------------+       +------------------+
                                    |
                                    v
                           +------------------+
                           |  Evidence Pack   |
                           |  (optional)      |
                           |  Shareable proof |
                           +------------------+

Supported input types:

  • Suspicious text from WhatsApp, SMS, email, Telegram, or social media
  • Screenshots of suspicious messages
  • QR codes and UPI payment requests
  • Payment confirmation screenshots (fake receipt detection)

Output for every scan:

  • Verdict: safe, risky, or unclear -- three states only, no ambiguity
  • Scam type classification
  • Plain-language reasoning
  • Confidence band
  • Recommended next action
  • Optional share shield and evidence pack

API

Chetana exposes a public v0 API for programmatic scam checks.

# Scan a suspicious message
curl -X POST https://chetana.activemirror.ai/api/v0/scan \
  -H "Content-Type: application/json" \
  -d '{
    "input_type": "text",
    "text": "Urgent: your bank account will be blocked today. Pay Rs 500 now.",
    "language_hint": "en",
    "session_id": "example-session"
  }'
Endpoint Purpose
POST /api/v0/scan Submit content for scam analysis
POST /api/v0/evidence Retrieve evidence pack for a scan
POST /api/v0/events Event logging

Architecture

Layer Technology
Frontend Vite 5, React 18, TypeScript, Tailwind CSS, Framer Motion
Backend FastAPI, Python 3.11, Pydantic
OCR Tesseract.js (client-side image text extraction)
On-device ML Hugging Face Transformers (browser inference)
Deployment FastAPI serves the built frontend as a single-origin application

Local Development

git clone https://github.com/MirrorDNA-Reflection-Protocol/chetana-site.git
cd chetana-site

# Frontend
cd frontend
npm install
npm run dev

# Backend (separate terminal)
cd backend
pip install -r requirements.txt
uvicorn app.main:app --port 8093

Design Principles

  • Three verdicts only. safe, risky, or unclear. No numeric risk scores that imply false precision.
  • Default to unclear. When evidence is weak, the system says so rather than guessing.
  • Show reasoning. Every verdict includes the factors that produced it.
  • Surface recovery rails. Official reporting channels are always visible, never buried.
  • No government affiliation. Chetana is an advisory tool. Verdicts are automated assessments, not legal determinations.

Recovery Resources

If you have been scammed:

  • National Cybercrime Helpline: 1930
  • File a report: cybercrime.gov.in
  • Women helpline: 181

Act within the first hour for the best chance of recovery.

Compliance

Chetana is built within the Active Mirror governed AI framework. See COMPLIANCE.md for control mappings against the EU AI Act, India DPDP Act 2023, SOC 2 Type II, and ISO 27001:2022.

Security

To report a vulnerability, see SECURITY.md. Do not use public GitHub issues for security reports.


Not affiliated with the Government of India, RBI, UIDAI, CERT-IN, or any law enforcement agency.

Built by Active Mirror -- governed AI for institutional work.

Releases

No releases published

Packages

 
 
 

Contributors