Indian Constitution RAG Model

This repository contains a retrieval-augmented generation chatbot for asking questions against a local PDF copy of the Constitution of India. It uses LangChain, Hugging Face sentence embeddings, FAISS vector search, Groq-hosted chat models, and Streamlit.

Value Proposition

The project demonstrates the core pieces of a RAG workflow:

• load and chunk a PDF
• build a local vector index
• retrieve relevant context
• generate a grounded answer
• show retrieved snippets for transparency
• optionally convert answers to audio

Architecture

flowchart LR
    A["CONSTITUTION.pdf"] --> B["PyPDFLoader"]
    B --> C["Text splitter"]
    C --> D["Hugging Face embeddings"]
    D --> E["FAISS index"]
    F["User question"] --> G["Retriever"]
    E --> G
    G --> H["Prompt with retrieved context"]
    H --> I["Groq chat model"]
    I --> J["Answer + source snippets"]
    J --> K["Streamlit UI"]

Loading

Project Structure

RAG_MODEL/
|-- main.py
|-- build_index.py
|-- requirements.txt
|-- README.md
|-- .gitignore
|-- data/
|   `-- CONSTITUTION.pdf
|-- notebooks/
|   `-- RAG_CONSTITUTION.ipynb
|-- src/
|   `-- constitution_rag/
|       |-- app.py
|       |-- audio.py
|       |-- indexing.py
|       |-- rag_chain.py
|       `-- settings.py
`-- tests/
    `-- test_settings.py

Quick Start

python -m venv .venv
.venv\\Scripts\\activate
pip install -r requirements.txt
python build_index.py

Create .streamlit/secrets.toml:

GROQ_API_KEY = "your_groq_api_key"
FAISS_INDEX_TRUSTED = "true"
ENABLE_AUDIO = "true"

Run the app:

streamlit run main.py

On macOS/Linux:

source .venv/bin/activate

Configuration

Setting	Required	Purpose
GROQ_API_KEY	Yes	API key for Groq chat completion
FAISS_INDEX_TRUSTED	Yes	Allows loading the local FAISS pickle index only after you confirm it was generated from trusted files
RAG_GROQ_MODEL	No	Overrides the default Groq model
ENABLE_AUDIO	No	Enables or disables gTTS audio output

Security Notes

• FAISS local indexes use pickle internally. This app blocks loading until FAISS_INDEX_TRUSTED=true is set.
• Only set FAISS_INDEX_TRUSTED=true for an index you generated yourself with build_index.py.
• Do not commit .env, .streamlit/secrets.toml, or API keys.
• The app is not legal advice. It answers from retrieved PDF context and may be incomplete.

Improvements Included

• Added reproducible build_index.py.
• Added guarded FAISS index loading instead of unconditional dangerous deserialization.
• Added source snippet display.
• Added environment/secrets-based configuration.
• Added cached embedding, LLM, and vector-store loading.
• Added clearer error handling for missing keys and missing indexes.

Development Workflow

set PYTHONPATH=src
python -m compileall main.py build_index.py src
pytest -q
python build_index.py
streamlit run main.py

On macOS/Linux, use export PYTHONPATH=src.

Roadmap

• Add citation formatting by page number
• Add retrieval evaluation set
• Add answer quality regression tests
• Add document upload support
• Replace single-document index with a managed document ingestion pipeline

FAQ

Does this provide legal advice? No. It is an educational RAG demo.

Why is FAISS_INDEX_TRUSTED required? LangChain FAISS indexes use pickle during local loading. Pickle can execute unsafe payloads, so this project requires explicit trust.

Why do answers mention limited context? The chatbot is instructed not to invent answers beyond retrieved source text.

Troubleshooting

Issue	Fix
FAISS index not found	Run python build_index.py.
API key error	Add GROQ_API_KEY to .streamlit/secrets.toml or environment variables.
FAISS trust error	Set FAISS_INDEX_TRUSTED=true after building the index locally.
Audio fails	Set ENABLE_AUDIO=false or check network access for gTTS.

License

No license file is currently included. Add a license before reusing or distributing this project.