Tathya (तथ्य)

AI-powered fact-checker for Instagram Reels & YouTube Shorts — in your language.

Tathya extracts audio from short-form videos, transcribes it, runs an LLM-based fact-check, and reads the verdict back to you via text-to-speech in 6 Indian languages.

📥 Download Android APK

---

How It Works

Paste URL → Extract Audio → Transcribe → Fact-Check → TTS Verdict

1. User pastes an Instagram Reel or YouTube Short URL (or shares directly from those apps)
2. Backend extracts audio via RapidAPI
3. Audio is transcribed using Groq Whisper
4. Claims are analyzed by Groq Llama 3.3 70B — returns verdict, confidence score, category, key points, and detailed reasoning
5. A spoken verdict is generated in the selected language using Sarvam AI TTS
6. Results are cached in MongoDB for instant repeat lookups

---

Tech Stack

Layer	Technology
Frontend	Expo (React Native) · TypeScript · Zustand · Expo Router
Backend	Python · FastAPI · Pydantic
Database	MongoDB (via Motor async driver)
Audio Extraction	RapidAPI (youtube-mp36 + instagram-reels-downloader-api)
Transcription	Groq Whisper API
Fact-Checking	Groq Llama 3.3 70B
Text-to-Speech	Sarvam AI bulbul:v2

---

Supported Languages

Language	Code
Hindi (हिंदी)	hi-IN
Tamil (தமிழ்)	ta-IN
Telugu (తెలుగు)	te-IN
Kannada (ಕನ್ನಡ)	kn-IN
Malayalam (മലയാളം)	ml-IN
Marathi (मराठी)	mr-IN

---

Project Structure

├── backend/
│   ├── server.py                 # FastAPI app — /api/check, /api/health
│   ├── Dockerfile                # Production container (Python 3.11 + ffmpeg)
│   ├── render.yaml               # Render deployment blueprint
│   ├── requirements.txt          # Full dev dependencies
│   └── requirements.deploy.txt   # Slim production dependencies
├── frontend/
│   ├── app/
│   │   ├── _layout.tsx           # Root layout (dark theme, share intent provider)
│   │   ├── index.tsx             # Home — URL input, language picker, share intent handler
│   │   └── result.tsx            # Result — verdict card, audio playback
│   ├── components/
│   │   ├── VerdictCard.tsx       # Verdict display with enhanced context fields
│   │   ├── LanguagePicker.tsx    # Language selection modal
│   │   └── LoadingOverlay.tsx    # Loading animation
│   ├── store/
│   │   └── useCheckStore.ts      # Zustand store — API calls, state management
│   ├── constants/
│   │   └── languages.ts          # Language codes and labels
│   └── plugins/
│       └── withShareIntent.js    # Expo config plugin for Android share intent
└── memory/
    └── PRD.md                    # Product requirements document

---

Getting Started

Prerequisites

• Node.js ≥ 20
• Python ≥ 3.11
• MongoDB (local or Atlas)
• API keys for: Groq, Sarvam AI, RapidAPI (subscribed to youtube-mp36 and instagram-reels-downloader-api)

Backend

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Create backend/.env:

GROQ_API_KEY=your_groq_key
SARVAM_API_KEY=your_sarvam_key
RAPIDAPI_KEY=your_rapidapi_key
MONGO_URL=mongodb://localhost:27017
DB_NAME=tathya_db

Run the server:

uvicorn server:app --host 0.0.0.0 --port 8001 --reload

The API will be available at http://localhost:8001. Health check: GET /api/health.

Frontend

cd frontend
npm install --legacy-peer-deps

Create frontend/.env:

EXPO_PUBLIC_BACKEND_URL=http://localhost:8001

Start the dev server:

npx expo start

Use the Expo Go app on your phone or press w to open the web version.

---

API

POST /api/check

Fact-check a video URL.

Request body:

{
  "url": "https://www.instagram.com/reel/ABC123/",
  "language_code": "hi-IN"
}

Response: verdict, confidence (0–100), category, key points, detailed reasoning, TTS audio (base64), and regional language translations.

GET /api/health

Returns server status.

---

Deployment

Backend (Render — Free Tier)

1. Push to GitHub
2. Create a Web Service on render.com
3. Set root directory to backend, runtime to Docker
4. Add environment variables (GROQ_API_KEY, SARVAM_API_KEY, RAPIDAPI_KEY, MONGO_URL, DB_NAME)

Free tier sleeps after 15 min of inactivity. First request after sleep takes ~30s.

Frontend (Cloudflare Pages — Free Tier)

1. Create a project on pages.cloudflare.com
2. Connect your GitHub repo, root directory frontend
3. Build command: npx expo export --platform web
4. Output directory: dist
5. Env vars: EXPO_PUBLIC_BACKEND_URL=<your-render-url>, NODE_VERSION=20

Android APK (EAS Build)

Pre-built APK: Download the latest APK here

To build a new APK yourself:

cd frontend
npx eas-cli login
npx eas-cli build --platform android --profile preview

Database (MongoDB Atlas — Free Tier)

Create a free M0 cluster at mongodb.com/atlas. The app uses a single collection (checks) indexed on cache_key (unique).

---

MongoDB Schema

Collection: checks

{
  "cache_key": "sha256(url:language_code)",
  "url": "https://...",
  "language_code": "hi-IN",
  "claim": "...",
  "verdict": "TRUE | FALSE | MISLEADING | PARTIALLY_TRUE",
  "confidence": 85,
  "reason": "...",
  "verdict_text": "...",
  "audio_base64": "...",
  "category": "health | science | history | technology | finance | news | general",
  "key_points": ["..."],
  "fact_details": "...",
  "what_to_know": "...",
  "sources_note": "...",
  "created_at": "2026-02-28T17:31:45.228Z"
}

---

License

MIT