Ascend AI Tutor 🚀

An immersive, production-grade AI Tutoring application designed specifically for Indian government exams (UPSC, SSC CGL, IBPS PO/SO/Clerk, RRB NTPC, State PSC).

🌟 Features

1. 3D Knowledge Universe

   • Built with @react-three/fiber and @react-three/drei.
   • Explore an interactive, floating node graph of the entire syllabus. Complete with custom shaders, particle systems, and glassmorphic tooltips.
   • Navigate to /explore to experience it.

2. Native Voice Interface

   • Seamless, zero-latency Web Speech API integration.
   • Beautiful, framer-motion powered microphone animations (concentric pulsing glow rings).
   • Zero additional API costs. Talk directly to your AI Tutor.

3. RAG-Augmented LLM Pipeline (Backend)

   • Features a robust RAG-augmented LLM pipeline: user question → ChromaDB retrieval → context injection → Llama 3.3 stream.

4. Progressive Web App (PWA)

   • Installable on mobile devices via Chrome/Safari.
   • manifest.json configured for a native-app-like experience.

5. Premium UI/UX & Glassmorphism

   • Complete visual overhaul using brand colors: Deep Space Background (#050810), Neon Cyan (#00D4FF), and Electric Violet (#8B5CF6).
   • Extensive use of Tailwind's backdrop-blur for heavy glassy overlays, creating incredible depth.

---

🛠 Tech Stack

• Frontend: Next.js 16 (App Router), React 19, Tailwind CSS v4, Zustand, Framer Motion, React Three Fiber.
• Backend: FastAPI, SQLAlchemy, PostgreSQL, Alembic, ChromaDB + fastembed (RAG), Redis, structlog, slowapi, Sentry.
• AI Models: Llama 3.3 via Groq API — streamed responses with SSE.

---

⚙️ Architecture

• RAG Pipeline: Retrieves factual content from a structured exam corpus (pyqs.json — UPSC, SSC CGL, IBPS PO/SO/Clerk, RRB NTPC, State PSC) via ChromaDB semantic search + fastembed embeddings.
• Response Generation: Llama 3.3 via Groq API generates streamed responses grounded in the retrieved context.
• Auth: httpOnly JWT cookies with 1-hour access tokens + 30-day refresh token rotation. CSRF protection via Origin/Referer validation.
• Error Handling: Sentry integration on both frontend (global-error.tsx) and backend (general_exception_handler). Structured JSON logging with X-Request-ID correlation.

---

📦 License

MIT

---

🚀 Getting Started

To spin up the entire stack locally using Docker:

cp frontend/.env.example frontend/.env.local
cp backend/.env.example backend/.env
docker compose up --build

This boots up the PostgreSQL database, the ChromaDB vector store, the FastAPI backend (and auto-ingests the pyqs.json data), and the Next.js frontend.

---

🚀 Deploy

See DEPLOY.md for Railway + Vercel.

What is still required before this is a shippable, paid product: PRODUCTION_READINESS.md.