🧾 AI Invoice Analyzer (FastAPI + LLM + Vision + PDF)

An AI-powered invoice and receipt analyzer that extracts structured invoice data from images (JPG/PNG) and PDF documents using a Large Language Model (LLM).

The system supports vision-based parsing, PDF text extraction, and outputs validated JSON via a FastAPI backend.

---

📹 Demo Video

Watch Demo Video

🚀 Features

• 📸 Parse invoice images (JPG, PNG)
• 📄 Parse text-based PDFs
• 🤖 Uses GitHub-hosted LLM via OpenAI-compatible SDK
• 🧠 Smart vendor detection (logos, headers, prominent text)
• 🧾 Extracts structured fields:
  • Vendor
  • Invoice number
  • Invoice date
  • Due date
  • Total amount
  • Currency
  • Validation flag
• 🌐 FastAPI REST API
• 🧩 Schema-safe & production-ready (no API crashes)

---

📁 Project Structure

ai-invoice-analyzer/
│
├── main.py              # FastAPI backend
├── requirements.txt     # Python dependencies
├── .env                 # Environment variables (not committed)
├── .gitignore
└── README.md

⚙️ Tech Stack

Backend: FastAPI, Pydantic
AI Model: GitHub-hosted LLM (OpenAI SDK compatible)
Vision: Image-based invoice parsing
PDF Parsing: PyMuPDF
Deployment-ready: Uvicorn

🔐 Environment Setup

Create a .env file in the project root:

BASE_URL=https://models.inference.ai.azure.com
API_KEY=your_github_model_api_key
MODEL_NAME=gpt-4o-mini

⚠️ Never commit .env to GitHub.

📦 Installation

pip install -r requirements.txt

Required packages:

fastapi
uvicorn
openai
python-dotenv
python-multipart
pymupdf

▶️ Run the Application

uvicorn main:app --reload

📤 API Usage

Upload Invoice (Image or PDF)

curl -X POST http://127.0.0.1:8000/analyze-invoice \
  -H "accept: application/json" \
  -F "file=@invoice.jpg"

📥 Example Response

{
  "vendor": "ABC Seller",
  "invoice_number": "INV-2025-019",
  "invoice_date": "2025-12-20",
  "due_date": null,
  "total_amount": "530.00",
  "currency": "BDT",
  "valid": true
}

🧠 How It Works

Image Invoices

1. Invoice image is base64 encoded.
2. Sent to a vision-capable LLM.
3. Vendor extracted from logo/header.
4. Structured JSON returned with fields:
   • vendor
   • invoice_number
   • invoice_date
   • due_date
   • total_amount
   • currency
   • valid

PDF Invoices

1. Text extracted using PyMuPDF.
2. Extracted text sent to LLM.
3. Invoice fields parsed from content into structured JSON.

🙌 Author

Md. Zobayer Ibna Kabir
CSE Graduate | AI & ML Enthusiast
GitHub: https://github.com/ZobayerAkib

⭐ If you find this project useful, consider giving it a star!