Google's File Search + SpineDAO HeritageNEt

Advanced OCR, Vector Search & Knowledge Graph Extraction for Historical Medical Documents

🚀 Features

Core Capabilities

• Advanced OCR Engine: Gemini AI-powered text extraction with image preprocessing
• Vector Search: Google File Search for semantic document retrieval
• Knowledge Graph: CAMEL-AI + Neo4j for entity extraction and relationship mapping
• RAG Support: Retrieval Augmented Generation for contextual answers

🏗️ Architecture

graph TD
    User([User]) -->|Upload PDF| Flask 
    
    subgraph "Pipeline A: Semantic Search"
        Flask -->|1. Extract Metadata| Meta[Metadata Extractor]
        Meta -->|2. Index Document| VectorDB[(☁️ Google Vector Store)]
        User -->|Query| VectorDB
    end
    
    subgraph "Pipeline B: Knowledge Graph"
        Flask -->|1. Vision OCR| OCR[👁️ OCR Engine]
        OCR -->|2. Raw Text| Agent[🤖 Camel-AI Agent]
        Agent -->|3. Extract Entities| GEMINI[GEMINI_2-0-FLASH]
        GEMINI -->|4. Commit Data| Neo4j[(🧠 Neo4j Graph DB)]
    end

Loading

📋 Prerequisites

• Python 3.11+
• Google Cloud Platform account (for Gemini AI)
• Neo4j database
• Groq account

🛠️ Installation

1. Clone the repository

   git clone <repository-url>
   cd file-search

2. Install dependencies

   uv sync

3. Set up environment variables Create a .env file in the project root:

   # Google AI Services
   GEMINI_API_KEY=your_gemini_api_key_here
   
   # Neo4j Database
   NEO4J_URI=bolt://localhost:7687
   NEO4J_USERNAME=neo4j
   NEO4J_PASSWORD=your_neo4j_password
   
   # Optional
   PORT=5000

🚀 Quick Start

1. Run the application

   uv run main.py

2. Open your browser Navigate to http://localhost:5000

📚 Usage

Web Interface

1. Upload Documents: Drag and drop PDF files onto the upload area
2. Ask Questions: Use the chat interface to query processed documents
3. View Library: Monitor processed documents in the sidebar
4. Delete Documents: Remove documents from the search index

OCR Engine Options

• Enhancement Levels: light, medium, aggressive
• DPI Settings: 200 or 300 for scan quality
• Medical Context: Specialized prompts for medical terminology
• Preprocessing: Image enhancement and deskewing

File Search Settings

• Chunk Size: 512 tokens per chunk
• Overlap: 50 tokens between chunks
• Model: Gemini-2.0-flash for search and responses

Knowledge Graph Schema

• Node Types:

  • ClinicalObservation
  • TherapeuticOutcome
  • ContextualFactor
  • MechanisticConcept
  • TherapeuticApproach
  • SourceText

• Relationship Types:

  • co_occurs_with
  • preceded_by/followed_by
  • modified_by
  • responds_to
  • associated_with
  • results_in
  • described_in
  • contradicts/corroborates

📁 Project Structure

├── main.py                 # Flask application entry point
├── file_search.py          # Vector search engine (Google AI)
├── ocr_engine.py           # Advanced OCR with Gemini Vision
├── kg_agents.py            # Knowledge graph extraction agent
├── templates/
│   └── index.html          # Web interface
├── uploads/                # Temporary file storage
├── debug_images/           # OCR preprocessing debug output
├── pyproject.toml          # Python dependencies
└── Procfile               # Heroku deployment config