LegalMind AI
A question-answering system over legal documents. Upload PDFs, ask in plain English, and get answers with citations to the exact passages they came from.
- Type
- AI, retrieval-augmented generation
- Status
- Open source
- Built with
- Python, LangChain, FAISS, HuggingFace embeddings, Groq, Streamlit
- Links
- View the code
The problem
Legal and religious texts are long, dense and full of cross-references. Finding the paragraph that answers a specific question means reading far more than you need to, and a general chatbot will happily invent an answer that sounds right.
What I built
A retrieval-augmented generation (RAG) system. Documents are split into chunks, turned into embeddings with HuggingFace sentence transformers and indexed in FAISS. When someone asks a question, the most relevant chunks are retrieved and passed to a fast LLM on Groq, which answers using only that material.
- Works with PDF and text files
- Every answer shows the source passages it was built from
- Choose which documents to search
- Adjustable chunk size, overlap and temperature
- A simple Streamlit interface, orchestrated with LangChain
Why it matters for clients
This is the same architecture behind the AI website assistant I build for businesses: answers that come from the business’s own content, with the source to prove it.