Gita AI Rag chat bot 🤖

An intelligent assistant that uses Google's Gemini AI to provide accurate answers about Gita.

System Architecture

graph TD
    A[Docs] --> B[Document Processor]
    B --> C[Text Chunks]
    C --> D[SentenceTransformer Embedding Model]
    D --> E[FAISS Vector Store]

    F[User Question] --> G[Question Embedding]
    G --> H[Similarity Search]
    E --> H
    H --> I[Context Retrieval]
    I --> J[Gemini Pro]
    J --> K[Generated Answer]

Loading

🚀 Quick Start

Using Docker

# Build the image
docker build -t crustdata-assistant .

# Run the container
docker run -p 8000:8000 \
  -e GEMINI_API_KEY=your_key_here \
  crustdata-assistant

Local Development Setup

# Create virtual environment
python -m venv venv

# Activate virtual environment
source venv/bin/activate  # Linux/Mac
# or
venv\Scripts\activate    # Windows

# Install dependencies
pip install -r requirements.txt

# Run the application
python main.py

📁 Project Structure

project/
├── data/
│   ├── processed/
│   │   ├── faiss_index/    # Vector embeddings
│   │   └── chunks/         # Processed text chunks
│   └── raw/                # Raw documentation
├── src/
│   ├── core/              # Core processing logic
│   ├── api/               # API endpoints
│   ├── services/          # Gemini integration
│   └── utils/             # Helper functions
├── Dockerfile
├── requirements.txt
└── main.py

🔄 Processing Pipeline

sequenceDiagram
    participant D as Documentation
    participant C as Chunker
    participant E as Embedder
    participant F as FAISS
    participant U as User
    participant G as Gemini

    D->>C: Split into chunks
    C->>E: Generate embeddings
    E->>F: Store vectors

    U->>E: Question
    E->>F: Search similar docs
    F->>G: Context + Question
    G->>U: Generated Answer

Loading

🛠️ Key Components

1. Document Processor

• Splits documentation into manageable chunks
• Maintains document structure
• Processes markdown formatting

2. Embedding System

• Uses Gemini for generating embeddings
• Converts text chunks to vectors
• Optimizes for technical content

3. Vector Store (FAISS)

• Fast similarity search
• In-memory vector storage
• Efficient retrieval

4. Response Generator

• Context-aware answers
• Technical accuracy
• Natural language responses

📡 API Endpoints

Ask Question

POST /api/chat
Content-Type: application/json

{
    "question": "Who is arjuna?",
    "max_context": 3  # Optional: Number of relevant chunks to use
}

Health Check

GET /health

🔧 Configuration

Environment Variables

GEMINI_API_KEY=

# Application Settings
APP_ENV="development"
DEBUG=True
PORT=8000


# Embedding Model Configuration
EMBEDDING_MODEL="all-mpnet-base-v2"

# Processing Configuration
CHUNK_SIZE=1000
CHUNK_OVERLAP=100
MAX_TOKENS=8192

Docker Configuration

FROM python:3.9-slim

COPY requirements.txt .
RUN pip3 install -r requirements.txt

COPY . .
RUN mkdir -p data/raw data/processed/faiss_index data/processed/chunks logs

CMD ["python", "main.py"]

📊 Performance Considerations

• Embedding Generation: ~100ms per chunk
• Search Latency: ~50ms
• Response Generation: ~500ms
• Memory Usage: ~500MB base + ~100MB per 1000 chunks

🔍 Monitoring

Monitor system health:

curl http://localhost:8000/health

🚨 Error Handling

• Documents missing/corrupted
• Embedding generation failures
• Search index errors
• Response generation timeout

🔐 Security

• Input sanitization
• Error message sanitization