A smart Streamlit web application that leverages the power of Meta LLaMA 3 Vision-Instruct to extract structured data from invoices in image or PDF format. It supports intelligent field extraction, invoice summarization, and export to PDF and JSON formats.
- ⚙️ Built in BIP Smart Everything program
- 📤 Upload PDF or Image (JPG/PNG) invoices
- 🔍 Automatically extract:
- Service Provider
- Customer Information
- Service Description
- Invoice Date
- Invoice Total
- 📄 Convert extracted summaries to PDF
- 🧾 Generate structured JSON output
- 💬 Ask natural language questions about uploaded documents
- 🧠 Powered by LLaMA 3 Vision-Instruct on Hugging Face
| Tool / Library | Description |
|---|---|
| Streamlit | Web UI framework for building interactive apps |
| Hugging Face Inference Endpoints | Run multimodal LLMs (e.g., LLaMA 3) via API |
| pdf2image | Convert PDF pages to high-resolution images |
| PyPDF2 | Read, split, and merge PDF files |
| Pillow (PIL) | Image processing library for Python |
| ReportLab | Create PDF documents with custom layout |
| python-docx | Create and edit Word (.docx) documents |
| docx2pdf | Convert Word documents (.docx) to PDF |
| python-dotenv | Manage environment variables using .env files |
| Meta LLaMA 3.2 Vision-Instruct | Multimodal LLM for document and image understanding |
- More of the imported/installed libraries can be found in the AI_Model_V2_Documentation.txt file
.jpg,.jpeg,.png(image-based invoices).pdf(multi-page supported)
- Adriana Pereira Bastos
- Anabela Oliveira Araújo
- Aleksander Sarnatskiy
- This project is licensed under the MIT License. Feel free to use, modify, and distribute.
Developed by a students in the BIP Smart Everything program. For feedback, improvements or collaborations, please reach out via GitHub.
/frontend_astro– Contains the Astro-based frontend for the websites./backend_model– Contains different versions of the prototype that handles the document invoice parsing (AI_Model_V3.py is the latest version)/general_files– More detailed documents about the program and installations/libraries used.