GS360 — The Open-Source UPSC Command Center

Democratizing UPSC preparation. Zero-cost AI tutoring. Plug-and-play knowledge. Built for aspirants, by aspirants.

---

🚀 The Vision

In India, UPSC coaching is a multi-crore industry that often prices out the most deserving candidates. GS360 is the first open-source, agent-native platform designed to break this barrier. We aren't just a notes-app; we are a specialized AI ecosystem that transforms raw study materials into an interactive, cited, and accurate learning experience.

🧠 Why GS360?

1. Agent-Native Tutoring (Powered by DeepTutor)

Unlike generic LLMs, GS360 uses a custom-patched DeepTutor engine with 16k+ lines of agent logic. It doesn't just "chat"; it teaches. It remembers your progress, identifies your weak spots in the syllabus, and provides answers grounded strictly in your provided content.

2. Plug-and-Play Content Packs

The "moat" of GS360 is its community. Anyone can create a Content Pack (NCERTs, Laxmikanth summaries, PYQ banks) by simply organizing files into folders. The system auto-ingests these into isolated vector indices, making them queryable in seconds.

3. Hardened RAG Accuracy

We treat accuracy as an engineering discipline. Every week, the engine is tested against a Golden Dataset of 200+ verified UPSC questions. Our 6-step de-risking strategy ensures zero hallucinations about critical constitutional or factual data.

Category	Day-1 Target	Week 12 Target
Factual Recall	75%	90%
Comprehension	65%	80%
Analytical	45%	65%

4. Security-First Multi-Tenancy

We take student data privacy seriously. GS360 implements:

• Index Isolation: Your personal notes and uploads use physically separate vector stores.
• Path Sanitization: Mandatory realpath validation to prevent path traversal.
• Audit Logging: Request-level authentication bound to your unique workspace.

---

🛠️ The Architecture

graph TD
    A[Next.js Frontend] --> B[Auth Layer / NextAuth]
    B --> C[Security Middleware]
    C --> D[User Namespace Manager]
    D --> E[DeepTutor Backend]
    E --> F[RAG Pipeline]
    F --> G[(Shared Content Packs)]
    F --> H[(Private User Vault)]
    E --> I[LLM Fallback Chain]
    I --> J[Gemini / DeepSeek / Ollama]

Loading

---

⚡ Quickstart

Infrastructure Requirements

• Hardware: 4GB+ RAM for local vector operations.
• Runtime: Python 3.11+ and Node.js 20+.
• Package managers: pip and npm.
• Optional: Docker + Docker Compose for containerized run.

Installation

1. Clone

   git clone https://github.com/JINA-CODE-SYSTEMS/GS360.git
   cd GS360

2. Backend dependencies

   cd gs360-live
   python -m venv .venv
   .venv\Scripts\activate
   pip install -r requirements.txt

3. Frontend dependencies

   cd web
   npm install

4. Environment files

• Create gs360-live/.env from gs360-live/.env.example
• Create gs360-live/web/.env.local and set NEXT_PUBLIC_API_BASE=http://localhost:8000

5. Run (local)

• Backend (from gs360-live/backend):

  python -m uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

• Frontend (from gs360-live/web):

  npm run dev

• Open http://localhost:3000

Dev Branch Local Run (Latest)

For active development on dev, run the GS360 backend and web app directly:

1. Start backend: python -m uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload from gs360-live/backend
2. Start frontend: npm run dev from gs360-live/web
3. Open: http://localhost:3000

Notes:

• Frontend API base should target http://localhost:8000 in gs360-live/web/.env.local.
• If you see Invalid token: Signature verification failed, clear stale local login state by reloading and logging in again.
• DeepTutor deep_solve and deep_research can require extra DeepTutor config assets; GS360 dev fallback now keeps these sections functional with direct LLM responses.

Docker Run

From repo root:

docker compose up --build

Expected ports:

• Frontend: 3000
• Backend: 8000

---

🔐 Environment Configuration

Required keys (gs360-live/.env)

• LLM_BINDING=openai
• LLM_MODEL=gpt-4o-mini
• LLM_API_KEY=<your-key>
• LLM_HOST=https://api.openai.com/v1

Recommended keys

• EMBEDDING_API_KEY=<your-key>
• GS360_SECRET_KEY=<long-random-secret>

Frontend (gs360-live/web/.env.local)

• NEXT_PUBLIC_API_BASE=http://localhost:8000

Security guidance:

• Never commit .env files.
• Rotate keys if exposed in logs/chat history.

---

🧩 Current API Surface (Backend)

Core endpoints:

• GET /health
• POST /auth/token
• POST /api/chat
• POST /api/solve
• POST /api/research
• POST /api/quiz
• GET /api/knowledge
• POST /api/knowledge
• POST /api/notes
• POST /api/eval/live
• GET /api/packs
• POST /api/guide/generate

Realtime:

• WS /ws/chat?token=<jwt>

---

🧪 Troubleshooting

1) Failed to fetch on login

Cause: frontend points to wrong backend port.

Fix:

• Set NEXT_PUBLIC_API_BASE=http://localhost:8000 in gs360-live/web/.env.local.
• Restart frontend dev server.

2) Invalid token: Signature verification failed

Cause: stale JWT in browser localStorage after backend secret/key updates.

Fix:

• Refresh page and login again.
• The frontend now auto-clears token on 401 and returns to login.

3) Next.js runtime chunk errors (example: Cannot find module './778.js')

Cause: stale .next build cache.

Fix:

cd gs360-live/web
rm -rf .next
npm run dev

4) uvicorn not found

Cause: missing backend dependencies or wrong interpreter.

Fix:

• Activate virtual env and install requirements:

cd gs360-live
.venv\Scripts\activate
pip install -r requirements.txt

5) DeepTutor capability-specific config errors (deep_solve, deep_research)

Cause: missing DeepTutor runtime config assets in local environment.

Current behavior:

• Chat uses DeepTutor turn runtime.
• Solve/Research routes use robust direct-LLM fallback in dev so UI remains usable.

---

📁 Repository Layout

• gs360-live/backend: FastAPI backend + DeepTutor bridge
• gs360-live/web: Next.js frontend
• gs360-live/core: shared domain/runtime helpers
• gs360-live/scripts: validation, ingest, and evaluation scripts
• demo-gs360: standalone demo UI assets
• docs: governance/content/plan documents

---

🔌 Content Ecosystem

The platform ships with a "Cold-Start" suite of content packs including:

• UPSC PYQs (2000–2025): ~2,500 questions with AI-verified explanations.
• NCERT Essentials: Summaries and MCQs for History, Polity, Geography, and Economy.
• PIB/Survey/Budget: Real-time summaries for current affairs.

Learn how to contribute your own packs in the Content Guide.

---

📅 Roadmap (16-Week Hardened Plan)

• Week 1-2: Phase 0 — Multi-tenancy security spike & Auth integration.
• Week 3-6: Phase 1 — Core AI suite (Notes, Quizzes, Flashcards) + RAG baseline.
• Week 7-10: Phase 2 — GS360 Design System & Command Center UI.
• Week 11-13: Phase 3 — Content creation & Domain expert review.
• Week 14-16: Phase 4 — Soft launch, Beta feedback, & Public release.

See the full Implementation Plan for technical deep-dives.

---

🤝 Community & Governance

GS360 is maintained by Jina Code Systems LLP under a BDFL model. We prioritize transparency, security, and student success above all else.

• Conduct: CODE_OF_CONDUCT.md
• Governance: GOVERNANCE.md
• Security: SECURITY.md

---

💰 Funding & Cost

GS360 is free forever. We use a tiered scaling model (Free Tier → Bootstrap → Sponsors) to ensure the platform remains accessible even as the community grows to 10k+ users.

---

Built with ❤️ in India by Jina Code Systems LLP.