DocTract - Document RAG Assistant

Privacy-first document processing system with local AI inference, vector search, and conversational interfaces for intelligent document analysis.

🎯 What This Project Does

• 📄 Document Processing - Upload and process PDFs with intelligent text extraction
• 🧠 Local AI Inference - Mistral 7B Instruct running locally for complete privacy
• 🔍 Vector Search - PostgreSQL + PGVector for semantic document retrieval
• 💬 Conversational Interface - Ask questions about documents in natural language
• 🔒 Privacy-First - All processing happens locally, no data leaves your system

🌐 Live Demo

📺 Demo Video

🚀 How to Use

Local Setup

# Navigate to project directory
cd ai-portfolio/projects/doctract

# Install dependencies
uv sync

# Set up PostgreSQL with PGVector
# (Ensure PostgreSQL is running with pgvector extension)

# Run the Streamlit app
uv run streamlit run src/doctract/rag/app.py

Using the App

1. Upload PDF - Drag and drop any PDF document
2. Ask Questions - Type natural language questions about the content
3. Get Answers - Receive contextual responses from local AI
4. Adjust Settings - Tune retrieval parameters for optimal results

🔧 Key Features

• Local AI Processing: Mistral 7B Instruct (Q4_K_M) via llama.cpp for complete privacy
• Vector Search: PostgreSQL + PGVector for semantic document retrieval
• Smart Chunking: Intelligent text splitting with configurable parameters
• Interactive Interface: Streamlit UI with real-time configuration
• Fast Retrieval: Sub-second response times with optimized search

🛠️ Tech Stack

AI/ML: Mistral 7B Instruct, HuggingFace Embeddings, llama.cpp Database: PostgreSQL, PGVector Frontend: Streamlit Processing: PyMuPDF, LlamaIndex Language: Python 3.11+

🤝 Skills Demonstrated

• RAG Systems & Vector Databases
• Local AI Deployment & Privacy-First Design
• Document Processing & Text Extraction
• Interactive UI Development
• System Architecture & Modular Design