SQORA

SQORA is an AI-powered competitive exam preparation platform for JEE/NEET aspirants. It combines a real-time AI mentor, auto-generated math and science animations, mock exams, contest features, and local text-to-speech into one learning workspace.

Features

Feature	Description
AI Mentor	Chat with a Gemini-powered tutor that explains concepts, solves doubts, and maintains context across turns
Manim Animations	AI explanations trigger auto-generated animated videos rendered with Manim
Mock Exams	Take timed exams with auto-grading and review
Contest Arena	Browse upcoming and past contests
Context Compaction	Rolling Gemini-based context summarization keeps long chats efficient
Text-to-Speech	Local TTS server delivers narrated audio explanations
Admin Panel	Configure mentor greetings, voice settings, and platform options
3D Landing Page	React Three Fiber-powered interactive landing page
Google Auth	Firebase Google Sign-In — unified sign-in/sign-up, no passwords
Multi-User Support	Fully isolated data per Firebase UID on the server
Token Auth	Firebase ID tokens verified server-side — strict data isolation

Architecture

┌──────────────────────────────┐       ┌─────────────────────────────────────┐
│        Vercel (CDN)          │       │         IITJ RAID Server            │
│                              │       │                                     │
│   React SPA (static build)   │──────▶│  FastAPI Backend  (port 8000)       │
│                              │       │  Manim Worker     (background)      │
│  VITE_API_URL = raid server  │       │  TTS Server       (port 8089)       │
└──────────────────────────────┘       └─────────────────────────────┬───────┘
                                                                     │
                                              ┌──────────────────────┘
                                              │  Firebase (Google Cloud)
                                              │    • Authentication (JWT)
                                              └──────────────────────

Services

Service	Stack	Deployment
Frontend	React 18, Vite, React Router, Three.js, KaTeX	Vercel
Backend API	FastAPI, Gemini, Firebase Auth	RAID Server
Manim Worker	Python, Manim 0.19, Gemini code gen	RAID Server
TTS Server	Python (Pocket-TTS)	RAID Server

User Data

user_data/               ← on RAID server
└── {firebase_uid}/
    ├── chat_history.json
    ├── ai_cache.json
    ├── video_cache.json
    ├── incoming_jobs/        ← Manim job queue
    └── rendered_videos/      ← generated MP4 files

---

Getting Started (Local Development)

Prerequisites

• Node.js 18+
• Python 3.12+
• ffmpeg (required by Manim)

1. Clone and Install

git clone https://github.com/devlup-labs/sqora.git
cd sqora

# Frontend
cd Frontend && npm install && cd ..

# Backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Configure Environment Variables

Copy the example and fill in your values:

cp .env.example .env

Edit .env:

GEMINI_API_KEY=your_api_key

VITE_FIREBASE_API_KEY=...
VITE_FIREBASE_AUTH_DOMAIN=...
VITE_FIREBASE_PROJECT_ID=...
VITE_FIREBASE_STORAGE_BUCKET=...
VITE_FIREBASE_MESSAGING_SENDER_ID=...
VITE_FIREBASE_APP_ID=...
VITE_FIREBASE_MEASUREMENT_ID=...

# For local dev: frontend hits backend on port 8000
VITE_API_URL=http://localhost:8000

3. Run Locally

# Terminal 1 — Frontend
cd Frontend && npm run dev

# Terminal 2 — Backend API
./start_backend.sh

# Terminal 3 — Manim Worker
./start_manim.sh

# Terminal 4 — TTS Server (optional)
./start_tts.sh

• Frontend: http://localhost:5173
• Backend: http://localhost:8000

---

Deploying to Vercel (Frontend Only)

1. Push to GitHub

git push origin main

2. Import to Vercel

1. Go to vercel.com/new
2. Import the sqora repository
3. Vercel reads vercel.json automatically — no extra config needed

3. Set Environment Variables in Vercel Dashboard

Under Project → Settings → Environment Variables, add:

Variable	Value
VITE_FIREBASE_API_KEY	From Firebase Console
VITE_FIREBASE_AUTH_DOMAIN	your-project.firebaseapp.com
VITE_FIREBASE_PROJECT_ID	Your project ID
VITE_FIREBASE_STORAGE_BUCKET	your-project.appspot.com
VITE_FIREBASE_MESSAGING_SENDER_ID	Sender ID
VITE_FIREBASE_APP_ID	App ID
VITE_FIREBASE_MEASUREMENT_ID	Measurement ID
VITE_API_URL	Public URL of your RAID server backend, e.g. https://sqora.iitj.ac.in

VITE_API_URL is required on Vercel — it tells the frontend where to send API requests.

4. Add Vercel Domain to Firebase

Firebase Console → Authentication → Authorized domains → Add your *.vercel.app URL.

---

Deploying the Backend on IITJ RAID Server

1. Clone and Install

git clone https://github.com/devlup-labs/sqora.git
cd sqora
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

2. Configure .env

Same as local dev, but without VITE_* variables (those are frontend-only):

GEMINI_API_KEY=your_key

3. Enable HTTPS + CORS via Nginx

The frontend is on Vercel and the backend is on the RAID server — this is a cross-origin setup. The backend already includes CORS middleware. Expose it through Nginx:

server {
    listen 80;
    server_name your-raid-domain-or-ip;

    location /api/ {
        proxy_pass http://localhost:8000;
        proxy_set_header Authorization $http_authorization;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        # Required for SSE streaming
        proxy_buffering off;
        proxy_cache off;
        proxy_read_timeout 300s;
        # Allow Vercel frontend
        add_header 'Access-Control-Allow-Origin' 'https://your-app.vercel.app' always;
        add_header 'Access-Control-Allow-Headers' 'Authorization, Content-Type' always;
    }
}

For HTTPS (strongly recommended): sudo certbot --nginx -d your-domain.com

4. Enable Full Token Verification

For production security, enable cryptographic token verification:

1. Firebase Console → Project Settings → Service Accounts → Generate new private key
2. Save as Unmute/firebase-service-account.json
   ⚠️ Never commit this file — it is in .gitignore
3. pip install firebase-admin
4. Restart the backend

5. Run Services

nohup ./start_backend.sh &
nohup ./start_manim.sh &
nohup ./start_tts.sh &   # optional

Or use tmux / systemd for persistent sessions.

---

Authentication

• Users authenticate with Google via Firebase — no passwords
• Firebase issues a short-lived JWT (ID token) on login
• Every API request carries it as Authorization: Bearer <token>
• Backend verifies the token and extracts uid to scope all data
• All data lives under user_data/{uid}/ — users can only access their own

---

API Reference

All chat and user endpoints require Authorization: Bearer <firebase_id_token>.

Method	Endpoint	Auth	Description
POST	/api/chat	✅	Send a message, get AI reply + video_id
GET	/api/chat/stream?message=...	✅	SSE stream of AI response tokens
GET	/api/users/{uid}/chat	✅	Fetch user's chat history
POST	/api/users/{uid}/chat	✅	Save chat history
GET	/api/users/{uid}/videos/{id}/status	✅	Check if a video is rendered
GET	/api/users/{uid}/videos/{id}/ready	✅ token param	SSE — fires ready when MP4 is done
GET	/api/users/{uid}/videos/{id}	✅ token param	Stream rendered .mp4 (byte-range)
GET	/api/contests	—	List contests
GET	/api/exams/{code}	—	Fetch exam questions
GET/PUT	/api/admin/config	—	Read/update platform settings

---

LLM Context Compaction

Long conversations are summarized before sending to Gemini:

1. Full history is always stored locally and shown in UI
2. LLM receives: compacted memory + recent turns + new message
3. When estimated token count exceeds trigger_tokens, old context is summarized via Gemini
4. Configure in Unmute/config.json → llm.context_compaction

---

Tech Stack

Layer	Technologies
Frontend	React 18, Vite, React Router, Three.js / React Three Fiber, KaTeX, react-markdown
Backend	FastAPI, Google Gemini (google-genai SDK), Firebase Admin SDK
Animation	Manim 0.19, Gemini code generation, ffmpeg
Auth	Firebase Authentication (Google Sign-In)
TTS	Pocket-TTS
Deployment	Vercel (frontend), IITJ RAID Server (backend + worker)

License

MIT