README.md

Document AI – Invoice Extraction System

IDFC GenAI Hackathon | Convolve 4.0

Project Overview

This project is an end-to-end Document AI pipeline designed to extract structured information from invoice images and PDFs.

Key Capabilities

OCR-based text extraction (Image + PDF)

Rule + fuzzy matching extractors

Model-based Horse Power inference

Stamp & Signature detection (YOLO – optional)

Confidence score recalibration

Streamlit-based UI demo

Batch & single-document inference support

Extracted Fields

Dealer Name

Model Name

Horse Power

Asset Cost

Stamp Presence + Bounding Box

Signature Presence + Bounding Box

Document Confidence Score

System Achitecture

The system follows an end-to-end, modular pipeline that ingests invoice images or PDFs, extracts multilingual text using OCR, converts it into structured JSON, and applies rule-based, fuzzy, and model-driven logic for accurate field extraction. Vision models detect stamps and signatures, while EDA and confidence calibration ensure reliable outputs. A Streamlit UI enables real-time single-document inference, with batch processing supported for offline evaluation.

UI Screenshot

Screenshot shows Streamlit UI with extracted fields & confidence score.

Demo Flow

Upload invoice image or PDF

Click Run Extraction

View extracted fields and confidence score

Output JSON generated at:

sample_output/result.json

Executable Usage

Single Document Inference (UI / API-like)

for image : python src/executable.py invoice.png

for pdf : python src/executable.py invoice.pdf

Note: Processes only the given document

Batch Mode (Offline / EDA)

python src/executable.py --batch data/train

Imp Note: Processes all images in folder Generates combined JSON output