https://github.com/invincible1602/document-extraction-agent
https://github.com/invincible1602/document-extraction-agent
Last synced: 10 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/invincible1602/document-extraction-agent
- Owner: Invincible1602
- Created: 2025-08-20T08:36:14.000Z (10 months ago)
- Default Branch: main
- Last Pushed: 2025-08-20T09:03:31.000Z (10 months ago)
- Last Synced: 2025-08-20T10:28:46.562Z (10 months ago)
- Language: Python
- Size: 11.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Document-Extraction-Agent
# 📄 Document Extraction Agent
An AI-powered Python project to **extract structured data from documents** such as invoices, medical bills, and prescriptions. The app provides **confidence scoring and validation** for each extracted field. It features a **Streamlit UI** and supports **Google Gemini** API for deployment and Hugging Face models for local testing.
---
## Features
- Upload documents in PDF, PNG, JPG, or JPEG format
- Automatic document type detection (Invoice, Medical Bill, Prescription)
- AI-powered field extraction using:
- **Google Gemini** (for deployment)
- **Hugging Face GPT-OSS-20B** (for local testing)
- Confidence scoring and validation rules for extracted data
- Download results as JSON
- Clean and interactive Streamlit UI
---
## Installation
1. Clone the repository:
```bash
git clone https://github.com/Invincible1602/Document-Extraction-Agent.git
cd Document-Extraction-Agent/app
```
## Project Structure
```bash
app/
├── main.py # Streamlit frontend
├── extraction.py # LLM-based extraction logic
├── agent_loop.py # Orchestrates document processing
├── document_processor.py # Handles PDF/image parsing
├── schemas.py # Pydantic models for structured output
├── confidence.py # Confidence scoring logic
└── requirements.txt
utils/
├── helpers.py # Text cleaning, date/amount validation, and other helpers
└── constants.py # Constants like supported file types, default fields
```