GenRagSystem (Shabdam)

An interactive, 3D digital experience powered by an AI-grounded Hybrid RAG (Retrieval-Augmented Generation) assistant for GenLab.

---

Tech Stack

Frontend (shabdham/)

• Framework: Nuxt 4 / Vue 3 (Composition API, <script setup>)
• 3D Graphics & Animations: Three.js / WebGL (interactive GLTF/GLB model rendering and canvas interactions)
• Language & Styling: TypeScript, Vanilla CSS (Design system with glassmorphism, smooth transitions, and responsive layouts)
• Streaming: Native Fetch Streams with custom Server-Sent Events (SSE) token queue consumer for typewriter-style real-time response rendering

Backend (GenRAG/)

• API Framework: FastAPI with async Server-Sent Events (StreamingResponse)
• Environment & Package Manager: uv (Python 3.10)
• LLM Inference: Groq API (openai/gpt-oss-20b) for ultra-low-latency intent routing and response generation
• Vector Database: Qdrant (Cloud/Local vector store for dense retrieval)
• Embeddings: Hugging Face Inference API
• Retrieval Engine: Hybrid Search combining:
  • Sparse: BM25 (rank-bm25)
  • Dense: Qdrant vector similarity
  • Fusion: Reciprocal Rank Fusion (RRF)
• Document Processing: pypdf (PDF corpus parsing & semantic chunking)
• Conversation Memory: SQLite for session-based chat history persistence

---

Quick Start

1. Backend Setup (GenRAG)

cd GenRAG
uv sync
uv run uvicorn api:app --reload --port 8000

2. Frontend Setup (shabdham)

cd shabdham
npm install
npm run dev

Visit http://localhost:3000 to run the app.