🏦 RBI NBFC Chatbot - Intelligent Regulatory Assistant

A sophisticated RAG-powered chatbot using Google Gemini AI, FAISS vector search, and 716 document chunks from official RBI guidelines.

---

🏗 Architecture

sequenceDiagram
    participant User
    participant Streamlit as Web UI
    participant App as FastAPI/App layer
    participant RAG as RAG Chain
    participant VectorDB as FAISS (Vector Store)
    participant LLM as Google Gemini

    User->>Streamlit: Asks question
    Streamlit->>App: Sends text
    App->>RAG: Invokes Chain
    RAG->>VectorDB: Query similar chunks (Top-4)
    VectorDB-->>RAG: Returns context
    RAG->>LLM: Prompt + Context + Question
    LLM-->>RAG: Generated Answer
    RAG-->>App: Answer + Source Metadata
    App-->>Streamlit: Display result
    Streamlit-->>User: Shows answer & sources

Loading

---

🚀 Quick Start

Option A: Docker (Recommended)

1. Build the image

   make docker-build
   # OR
   docker build -t rbi-bot .

2. Run the container

   # Ensure .env exists with GOOGLE_API_KEY
   make docker-run
   # OR
   docker run -p 8501:8501 --env-file .env rbi-bot

3. Access Open http://localhost:8501

Option B: Local Development

1. Setup

   make install
   cp .env.example .env  # Add your GOOGLE_API_KEY

2. Run

   make run

---

📋 What This Does

✨ 330-page RBI Master Direction → 716 smart chunks → Accurate answers with citations

Interfaces Available

Interface	Command	Use Case
🌐 Web UI	make run	Best for demonstrations
💬 Interactive CLI	python examples/demo_interactive.py	Quick Q&A sessions
🔌 REST API	python examples/demo_api.py	Integration, automation

---

🛠 Development

This project uses modern Python tooling.

# Linting (Ruff + Mypy)
make lint

# Testing (Pytest)
make test

Technical Stack

• LLM: Google Gemini 2.5 Flash
• Embeddings: Google text-embedding-004 (768-dim)
• Vector DB: FAISS (716 chunks)
• Framework: LangChain 0.2.16
• Container: Docker Multi-stage Build

---

📁 Project Structure

chatbot-langchain/
├── .github/              # CI/CD Workflows
├── src/                  # Source code
│   └── rbi_nbfc_chatbot/
│       ├── chains/       # RAG pipeline logic
│       ├── utils/        # Document processing
│       └── api/          # FastAPI server endpoints
├── tests/                # Pytest suite
├── data/                 # Vector store & documents
├── Dockerfile            # Container definition
├── Makefile              # Task runner
└── pyproject.toml        # Tool configuration

---

📈 Performance & Specs

• ⚡ 2-5 seconds per query
• 🎯 716 chunks searchable
• 🔍 Top-4 document retrieval
• 💾 ~600MB Docker Image size (Optimized)

---

Built with ❤️ using Google Gemini, LangChain, and FAISS