DocuMind - Intelligent Document Chat

DocuMind is an advanced RAG (Retrieval-Augmented Generation) application that allows users to seamlessly interact with their documents using AI. By leveraging the power of Google's Gemini models, Groq's high-speed inference, and vector search, DocuMind provides accurate, context-aware answers from your uploaded files and databases.

Live Demo

• Frontend Application: https://querywise.vercel.app
• Backend API: https://documind-p046.onrender.com

Modular RAG Architecture

DocuMind is built upon a highly scalable Modular RAG framework, ensuring precise context retrieval and coherent responses. The system comprises 6 Main Components:

1. Orchestrator (RAG Service): Coordinates the entire workflow, managing the flow of data between user input, retrieval, and generation.
2. Document Ingestion Engine: Handles the parsing, cleaning, and segmentation of various file formats (PDF, DOCX, TXT).
3. Embedding Service: Transforms text chunks into high-dimensional vector embeddings for semantic understanding.
4. Vector Store (Pinecone): Manages high-performance similarity search and retrieval of context.
5. Parent-Child Indexing Service: Implements advanced indexing strategies to maximize retrieval quality.
6. Generation Service (Gemini/Groq): Leverages state-of-the-art LLMs to synthesize answers using the retrieved context.

Vector Storing Methods: Parent & Child Indexing

To overcome the limitations of standard chunking, DocuMind employs a Parent-Child Indexing strategy:

• Child Chunks: Small, dense text segments (~300 chars) responsible for high-accuracy semantic search.
• Parent Documents: Larger context blocks linked to the child chunks.
• Retrieval Logic: When a child chunk matches a user's query, the system retrieves its corresponding Parent Document. This ensures the LLM receives full, coherent context rather than fragmented snippets.

Key Features

🧠 Advanced RAG Engine

• Chat with Documents: Upload PDFs, DOCX, or text files and ask questions in natural language.
• Transparent AI: View Vector Scores (V) and Resonance Scores (R) to understand exactly why a document was retrieved.
• Strict Scoring: Uses a local Reranker (ms-marco-MiniLM-L-12-v2) to prioritize accuracy over recall.
• HyDE: Uses Hypothetical Document Embeddings to improve search relevance.

📊 Database Intelligence & Visualization

• Text-to-SQL: Connect your database and ask questions in plain English (e.g., "Show me total sales by region").
• Auto-Visualization: The system automatically detects data patterns and generates Bar, Line, or Pie charts instantly.
• Schema Awareness: Automatically extracts and understands your database structure for accurate queries.
• SQL Safety: Built-in validator prevents destructive queries and ensures syntax correctness.

⚡ Performance & Security

• Multi-Model Support: Switch between Google Gemini 2.5 for reasoning and Groq (Llama 3) for ultra-fast responses.
• Secure Authentication: Robust user management with unique user IDs and secure session handling.
• Smart History: Persistent chat sessions allowing you to revisit previous conversations.
• Mobile Responsive: A modern, mobile-friendly interface built with React and Tailwind CSS.

Technology Stack

• Frontend: React, Vite, Tailwind CSS, Recharts
• Backend: Python, FastAPI
• AI/LLM: Google Gemini, Groq
• Vector DB: Pinecone
• Database: MongoDB Atlas (Chat History), PostgreSQL/MySQL (Data Analysis)
• Deployment: Vercel (Frontend), Render (Backend)