Aibys Invoice Extractor

AI-powered invoice and receipt data extractor. Upload PDFs or images, extract structured data with Ollama, then keep the results in local JSON and CSV files for ongoing use.

Part of the Aibys Document Intelligence series.

Features

• Upload multiple PDF, JPG, PNG, or WEBP files in one batch
• PDF pages are converted to images and extracted page by page
• AI extracts invoice data such as vendor, customer, line items, totals, dates, and payment method
• Structured result view with saved extraction history
• Raw JSON output for each record
• Persistent local storage using plain files, no SQL database
• CSV is appended to the same data/invoices.csv file across sessions and days
• Download saved CSV or JSON database from the UI
• Fully local when using a local Ollama vision model
• Responsive UI with no frontend build step

How It Works

1. Upload one or many invoices or receipts.
2. FastAPI validates each file and stores the original upload in uploads/.
3. PDF files are rendered into page images with PyMuPDF.
4. Each image is sent to Ollama with a structured extraction prompt.
5. Extracted records are saved into data/invoices.json.
6. Flattened rows are appended into data/invoices.csv.
7. The UI renders the latest batch and the saved history.

Quick Start

Prerequisites

• Python 3.10+
• Ollama running locally
• A vision-capable model, for example:

ollama pull gemma4:31b-cloud

Install & Run

git clone https://github.com/Arlchoose-code/aibys-invoice-extractor.git
cd aibys-invoice-extractor
pip install -r requirements.txt
uvicorn main:app --reload

Open:

http://localhost:8000

Configuration

Set environment variables as needed:

OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=gemma4:31b-cloud

Other possible vision models: llava, moondream, minicpm-v.

Local Data Files

The app creates these files and folders automatically:

Path	Purpose
uploads/	Original uploaded files
data/invoices.json	Full structured extraction history
data/invoices.csv	Append-only CSV export for spreadsheet use

This project intentionally uses JSON and CSV files instead of SQL so it stays simple and portable.

API

Method	Endpoint	Description
GET	/	Web UI
GET	/health	Ollama connection status
POST	/extract	Extract one or more files using multipart field files
GET	/records	Return saved JSON records
GET	/export-csv	Download saved CSV
POST	/export-csv	Export posted extraction data as CSV
GET	/export-json	Download saved JSON database

Tech Stack

• Backend: FastAPI + Python
• AI: Ollama
• PDF/Image Processing: PyMuPDF
• Storage: JSON + CSV files
• Frontend: Vanilla HTML/CSS/JS

Extracted Fields

Category	Fields
Document Info	Invoice number, date, due date, payment method
Vendor	Name, address, phone, email, website
Customer	Name, address, phone, email
Line Items	Description, quantity, unit price, total
Summary	Subtotal, tax, discount, shipping, total, currency

Aibys Document Intelligence Series

Repo	Description
Aibys Invoice Extractor	Extract invoice and receipt data
Aibys Legal Analyzer	Highlight risky clauses in contracts
Aibys Medical Explainer	Explain medical reports in plain language
Aibys Research Summarizer	Summarize academic papers

Author

Syahril Haryono - github.com/Arlchoose-code

License

MIT License