Folio — Document Intelligence System

A RAG-based document intelligence system that lets you have a real conversation with any document you upload.

What it does

Upload a PDF, ask questions in natural language, and get precise, source-grounded answers — no hallucinations, just what's actually in the document. Supports multi-turn conversation so you can ask follow-up questions naturally.

Tech Stack

• Frontend — Streamlit (dark/light theme)
• Embeddings — HuggingFace all-MiniLM-L6-v2 (384-dim)
• Vector Store — ChromaDB (local persistent)
• Retrieval — Maximal Marginal Relevance (MMR)
• LLM Backends — Groq (LLaMA 3.1-8B), Gemini 2.5 Flash, Mistral 7B
• Framework — LangChain

Features

• Multi-turn conversational QA with conversation memory
• MMR retrieval for diverse, non-redundant context
• Confidence scoring with visual indicator
• Readability analysis (Flesch Reading Ease)
• Auto-generated document summary
• Answer pinning and session export
• Persistent vector index (index once, reuse across sessions)

Getting Started

Prerequisites

• Python 3.10+
• API keys for your preferred LLM backend (Groq / Gemini / Mistral)

Installation

git clone https://github.com/tahirshamim/folio
cd folio
pip install -r requirements.txt

Setup

Create a .env file in the root directory:

GROQ_API_KEY=your_groq_key
GOOGLE_API_KEY=your_gemini_key
MISTRAL_API_KEY=your_mistral_key

Run

streamlit run app.py

How it works

1. Ingest — Document is parsed and split into 1000-char chunks with 200-char overlap
2. Embed — Each chunk is embedded using all-MiniLM-L6-v2 and stored in ChromaDB
3. Retrieve — Query is embedded and MMR search fetches the 5 most relevant, diverse passages
4. Generate — Retrieved passages + conversation history are sent to the selected LLM with a strict grounding prompt

Project Structure

folio/
├── app.py              # Main Streamlit application
├── requirements.txt
├── .env.example
└── chroma_store/       # Persistent vector index (auto-generated)

Roadmap

• Multi-document RAG
• Cross-encoder re-ranking
• REST API mode (FastAPI)
• RAGAS evaluation dashboard
• User authentication

Team

• Tahir Bin Shamim

License

MIT