AI Invoice Data Extractor Script

I need a browser-based script that lets me drop a PDF, JPG or PNG invoice into a simple web form and instantly returns a clean JSON payload with all key details. Accuracy is critical: the extraction must be reliable enough for downstream accounting automation, so please combine smart template-free parsing with solid OCR to reach a very high recognition rate.

Scope of the data to pull
• Seller information – company name, address, phone, VAT/Tax ID, plus any email or website you can confidently detect.
• Line-item table – product or service description, quantity, unit price (HT), VAT rate, and each row’s totals.
• Summary – subtotal, discounts, total VAT, grand total (TTC), currency.

Key constraints
• Language: invoices are primarily in French, so numeric formats (comma decimals, € symbol, “TTC”, etc.) must be handled gracefully.
• Input channel: files arrive only through an upload field on the page or API.
• Handle invoices with different layouts (not only one fixed template).
• Output: a single well-structured JSON object I can consume directly in my codebase.

Deliverables
1. Web-based script (can be: PHP, Python (Flask/Django/FastAPI), or Node.js – I’m open to suggestions.
2. Minimal HTML upload form and a results screen showing the raw JSON.
3. README that explains any third-party libraries, model training steps (if used), and how to extend to additional languages later.
4. Test run on a small set of sample French invoices I’ll provide; success is 95 %+ correct field population across that set.

Feel free to leverage tools like Tesseract, pdfplumber, PyMuPDF, PaddleOCR, spaCy, or even lightweight deep-learning models—whatever gets us the accuracy we need while keeping deployment straightforward. I’m happy to answer technical questions quickly so you can focus on clean code and solid extraction logic.