🎓 EduRAG Voice Assistant

An AI-powered educational voice assistant that answers questions from your knowledge base using Retrieval-Augmented Generation.

---

🌟 What is this?

EduRAG ingests your educational documents (NCERT textbooks, notes, PDFs-to-text, etc.), creates semantic embeddings, and lets you ask questions by voice. A LiveKit-powered voice agent retrieves the most relevant passages and answers in natural language — in English or Hindi/English bilingual mode.

Key Features

Feature	Description
🔍 Semantic Search	Find answers by meaning, not just keywords — powered by all-MiniLM-L6-v2 sentence-transformers
🎤 Voice Agent	Talk to your knowledge base using LiveKit real-time voice infrastructure
🌐 Bilingual	English-only or Hindi/English code-switching agent
📄 Auto-Ingest	Drop .txt, .md, .py, .json, or .csv files into documents/ and ingest with one command
🔒 Change Detection	MD5 hashing skips re-processing unchanged files
💾 Local-first	SQLite database — no external vector DB required

---

🏗️ Architecture

graph LR
    subgraph Client
        A[🎤 User Voice]
    end

    subgraph LiveKit Cloud
        B[WebRTC Room]
    end

    subgraph Voice Agent
        C[English Agent]
        D[Hindi/English Agent]
    end

    subgraph RAG Engine
        E[Embeddings<br/>all-MiniLM-L6-v2]
        F[Text Chunking<br/>Sentence Boundaries]
        G[SQLite Database]
    end

    subgraph AI Services
        H[Google Gemini<br/>LLM]
        I[Deepgram<br/>STT]
        J[Cartesia<br/>English TTS]
        K[Sarvam AI<br/>Hindi STT & TTS]
    end

    A <-->|WebRTC| B
    B <--> C
    B <--> D
    C --> E
    D --> E
    E <--> G
    F --> G
    C <--> H
    D <--> H
    C <--> I
    C <--> J
    D <--> K

Loading

Tech Stack: Python 3.10+ · LiveKit Agents · Sentence-Transformers · Deepgram · Google Gemini · Cartesia TTS · Sarvam AI · SQLite

---

🤖 AI Service Providers

The project uses a modular provider architecture. Here are the current and alternative options:

LLM (Large Language Model)

Provider	Model	Used In	API Key Env
Google Gemini ✅	gemini-2.5-flash	Both agents	GOOGLE_API_KEY
Groq	llama-3.3-70b, mixtral-8x7b	— (alternative)	GROQ_API_KEY
OpenAI	gpt-4o, gpt-4o-mini	— (alternative)	OPENAI_API_KEY
Anthropic	claude-sonnet-4-20250514	— (alternative)	ANTHROPIC_API_KEY

STT (Speech-to-Text)

Provider	Model	Used In	API Key Env
Groq ✅	whisper-large-v3-turbo	English agent	GROQ_API_KEY
Sarvam AI ✅	saarika:v2.5	Hindi agent	SARVAM_API_KEY
Deepgram	nova-2	— (alternative)	DEEPGRAM_API_KEY
Google Cloud	chirp	— (alternative)	GOOGLE_API_KEY

TTS (Text-to-Speech)

Provider	Model / Voice	Used In	API Key Env
Deepgram ✅	aura-2-thalia-en	English agent	DEEPGRAM_API_KEY
Sarvam AI ✅	anushka (Hindi)	Hindi agent	SARVAM_API_KEY
Cartesia	sonic-2	— (alternative)	CARTESIA_API_KEY
ElevenLabs	eleven_multilingual_v2	— (alternative)	ELEVENLABS_API_KEY

VAD (Voice Activity Detection)

Provider	Model	Used In
Silero ✅	silero-vad	Both agents (runs locally, no API key needed)

✅ = Currently configured in the codebase. To switch providers, update the entrypoint() function in the respective agent file.

---

📦 Project Structure

Rag-Mcp/
├── rag_mcp/                  # Core Python package
│   ├── config.py             #   Centralized config (frozen dataclass)
│   ├── database.py           #   SQLite layer (context-managed)
│   ├── embeddings.py         #   Embedding model (lazy-loaded)
│   ├── chunking.py           #   Sentence-boundary text chunking
│   ├── rag_engine.py         #   MCPRAGTool — high-level RAG API
│   ├── cli.py                #   CLI (ingest / search / stats / list)
│   └── agents/               #   LiveKit voice agents
│       ├── base.py           #     Shared logic + search tool
│       ├── english_agent.py  #     English-only agent
│       └── multi_agent.py    #     Hindi/English bilingual agent
├── documents/                # Knowledge base (NCERT textbooks)
├── voice-agent.py            # ★ Primary entry point
├── pyproject.toml            # Project metadata & tool config
├── requirements.txt          # pip dependencies
├── .env.example              # API key template
└── .gitignore

Note

The documents/ folder currently contains textbooks from NCERT (National Council of Educational Research and Training) covering subjects like Science, Maths, History, Economics, English Literature, Business Studies, and more. You can replace or add your own .txt, .md, or .csv files to build a custom knowledge base.

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• API keys for: LiveKit, Deepgram, Google AI, and optionally Cartesia, Sarvam AI, Groq

Installation

# 1. Clone the repository
git clone https://github.com/0xMihirK/EduRAG.git
cd EduRAG

# 2. Create a virtual environment
python -m venv .venv
.venv\Scripts\activate        # Windows
# source .venv/bin/activate   # macOS / Linux

# 3. Install dependencies
pip install -r requirements.txt

# 4. Download required model files
python voice-agent.py download-files

# 5. Set up API keys
copy .env.example .env        # Windows
# cp .env.example .env        # macOS / Linux
# Then edit .env with your actual keys

---

📖 Usage

🎤 Voice Agent

Start the voice assistant with LiveKit:

# Download model files (first time only)
python voice-agent.py download-files

# English agent (default)
python voice-agent.py dev

# Hindi / English bilingual agent
python voice-agent.py --language hindi dev

Then connect to the LiveKit room via the LiveKit Playground or your own frontend.

🔧 CLI — Document Management

# Ingest all documents from the documents/ folder
python -m rag_mcp --action ingest

# Semantic search
python -m rag_mcp --action search --query "Explain photosynthesis"

# Database statistics
python -m rag_mcp --action stats

# List ingested documents
python -m rag_mcp --action list

---

⚙️ Configuration

All settings can be overridden via environment variables:

Setting	Env Variable	Default
Database path	RAG_DB_PATH	rag_database.db
Embedding model	RAG_MODEL_NAME	all-MiniLM-L6-v2
Documents dir	RAG_DOCUMENTS_DIR	documents
Chunk size	RAG_CHUNK_SIZE	500
Chunk overlap	RAG_CHUNK_OVERLAP	50
Similarity threshold	RAG_SIMILARITY_THRESHOLD	0.3
Top-K results	RAG_TOP_K	5

See .env.example for the full list of required API keys.

---

🤝 Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit your changes: git commit -m "Add amazing feature"
4. Push to the branch: git push origin feature/amazing-feature
5. Open a Pull Request

---

🙏 Acknowledgements

• Sarvam AI — Hindi speech-to-text (saarika:v2.5) and text-to-speech models powering the bilingual agent
• LiveKit — Real-time voice infrastructure
• Deepgram — English speech-to-text and text-to-speech (aura-2)
• Google Gemini — Large language model for response generation
• Groq — Ultra-fast Whisper STT inference
• Sentence-Transformers — Semantic embedding models

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

---

Built with ❤️ for education