🎓 VIDYA OS 2.0

Sovereign Multi-Agent Campus Intelligence Platform

The "Bharat-First" intelligence platform for the global stage — sovereign, on-device, and running entirely on AMD hardware.

Offline Intelligence. Zero Cloud Fees. Absolute Data Sovereignty.

🧠 LEARN 2.0 · ⚙️ OPERATE 2.0 · 🎨 CREATE 2.0 · 🚀 Quick Start

---

🎯 The Global-Bharat Vision

VIDYA OS 2.0 is not just a hackathon project; it's a blueprint for Sovereign Digital Infrastructure. It solves the specific challenges of 40M+ Indian students while setting a global standard for decentralized AI.

• Indic-Language RAG 2.0: Native support for Hinglish, Tamil, and Bengali academic documents.
• Sovereign Compute: Optimized for AMD ROCm, leveraging local SLMs (Phi-4/Gemma 2) for enterprise performance without cloud costs.
• Liquid Intelligence UI: A premium, motion-first experience powered by Framer Motion.
• Agentic Orchestration: Moving beyond chatbots to a multi-agent system using LangGraph.

---

✨ Intelligence Pillars 2.0

🧠 LEARN — Indic AI Tutor

Sovereign academic coaching with multi-dialect support.

• On-Device SLM Inference: Powered by Phi-4/Gemma 2 via optimized Transformers.
• Indic-Language RAG: Query English textbooks using Hindi/Tamil voice commands.
• Concept Handover: The LEARN agent can "hand over" tasks to the CREATE agent to build practice tools.
• Zero-Latency Streaming: "Liquid" chat interface with real-time source attribution.

⚙️ OPERATE — Campus OS

Real-time intelligence dashboard with Bento-Grid evolution.

• Self-Organizing Bento Tiles: Layout shifts priority based on AI anomaly detection.
• Edge Analytics: Real-time footprints, energy, and space utilization processed locally.
• Neural Alerts: Natural language remediation guidance for campus emergencies.
• Digital Twin Transitions: Shared element transitions for building-level deep dives.

🎨 CREATE — Agentic Builder

Let any student build and deploy agentic tools — zero code required.

• Multi-Agent Templates: FAQ Agents, Campus Event Coordinators, Research Assistants.
• One-Click Local Deploy: Instant deployment to the sovereign campus network.
• Sovereign Embedding Store: Every apps leverages the campus's shared vector knowledge.

---

🏗️ Technical Architecture

graph TD
    subgraph "Frontend: Liquid UI (React + Framer Motion)"
        A[Landing Page] --> B[Indic Tutor]
        A --> C[Campus Bento OS]
        A --> D[Agentic Builder]
    end

    subgraph "Backend: Agentic Engine (FastAPI + LangGraph)"
        E[API Gateway] --> F[Coordinator Agent]
        F --> G[RAG Agent - Qdrant]
        F --> H[Tool Agent - SQL/Local]
        F --> I[Inference Service - ROCm]
    end

    subgraph "Sovereign Layer (AMD Hardware)"
        J[AMD Radeon/EPYC] --> K[ROCm Stack]
        K --> L[Local SLM: Phi-4/Gemma 2]
    end

    B <--> E
    C <--> E
    D <--> E

Loading

Global-Standard Tech Stack:

Layer	Technology
Core UI	React 19, Framer Motion, Lucide, Tailwind
Orchestration	LangGraph, LangChain, Pydantic AI
Inference	Transformers (ROCm Optimized), vLLM-lite
Vector DB	Qdrant (Persistent Collections)
Hardare Accel	AMD ROCm 6.2, Ryzen AI, EPYC

---

🚀 Quick Start

Prerequisites

Tool	Version	Install
Python	3.10+	python.org
Node.js	18+	nodejs.org
Ollama	Latest	ollama.com

1. Pull AI Models

ollama pull llama3.2

2. Clone & Start Backend

git clone https://github.com/archittmittal/Slingshot
cd Slingshot

# Windows (PowerShell)
.\start-backend.ps1

Or manually:

cd backend
python -m venv venv
# Windows:
venv\Scripts\activate
# macOS/Linux:
source venv/bin/activate

pip install -r requirements.txt
uvicorn main:app --reload --port 8000

3. Start Frontend

# Windows
.\start-frontend.ps1

Or:

cd frontend
npm install
npm run dev

4. Open in Browser

http://localhost:5173

---

🔌 API Reference

Method	Endpoint	Description
GET	/api/health	Health check + model info
POST	/api/learn/chat	Streaming LLM chat (SSE)
POST	/api/learn/quiz	Generate MCQ quiz from topic
GET	/api/operate/metrics	Live campus snapshot
GET	/api/operate/history	24-hour historical data
WS	/ws/campus	Live WebSocket feed (5s interval)
GET	/api/create/templates	List app templates
POST	/api/create/apps	Publish a new app
GET	/api/create/apps	List all published apps
POST	/api/create/apps/{id}/chat	Chat with a deployed app

Interactive API docs: http://localhost:8000/docs

---

🔧 AMD Deployment

VIDYA OS is architected to run on AMD hardware with zero code changes:

Dev (NVIDIA)	Production (AMD)
Ollama with CUDA	Ollama with ROCm
faster-whisper CUDA	faster-whisper ROCm
Any CUDA GPU	AMD Radeon RX 7900 / EPYC

Enable AMD ROCm:

# Linux — ROCm-enabled Ollama
OLLAMA_ROCm=1 ollama serve

# Or use the ROCm-specific Ollama build:
# https://ollama.com/download/linux#rocm

Every inference call stays on campus — no cloud, no cost per token, no data exposure.

---

📁 Project Structure

Slingshot/
├── backend/
│   ├── main.py              # FastAPI app — all APIs in one file
│   ├── requirements.txt
│   └── .env.example         # Copy to .env and configure
├── frontend/
│   └── src/
│       ├── App.jsx           # Root layout, sidebar, routing
│       ├── App.css
│       ├── index.css         # Global design system
│       └── pages/
│           ├── LandingPage.jsx / .css   # Hero, stats, pillar overview
│           ├── LearnPage.jsx / .css     # AI tutor + voice + quiz
│           ├── OperatePage.jsx / .css   # Live campus dashboard
│           └── CreatePage.jsx / .css    # No-code app builder
├── start-backend.ps1        # One-click backend startup (Windows)
├── start-frontend.ps1       # One-click frontend startup (Windows)
└── README.md

---

🌍 Responsible AI

Risk	Mitigation Built In
Data privacy	Zero data leaves campus — no cloud API
Hallucinations	Source attribution on every AI response
Bias in feedback	Structured, rubric-based evaluation
Content safety	Llama Guard filter layer
Misuse of app builder	Admin approval flow for published apps
Offline access	Fully functional without internet

---

🏆 Why This Wins — AMD Slingshot 2025

Judging Criteria	VIDYA OS
Innovation	First sovereign campus AI OS — not a chatbot wrapper
AMD Hardware	Every inference layer uses AMD silicon (ROCm + NPU)
Impact	40M+ Indian college students; works offline
Technical Depth	Streaming LLM, WebSockets, RAG-ready, multi-agent
Responsible AI	Privacy-by-design, explainability, content filters
Scalability	One AMD server → entire campus; open-source → any college

Cross-track coverage: GenAI (Track 1) + Education (Track 2) + Smart Cities (Track 8) + Social Good (Track 5) + Productivity (Track 3) + Cybersecurity (Track 6) → Open Innovation (Track 9)

---

👥 Team

Built for AMD Slingshot 2025 — National Hackathon

---

Built with ❤️ for India's 40 million college students

VIDYA OS · AMD Slingshot 2025 · Open Innovation