# Try AI document data extraction

> Paste the text of an invoice, a contract or a delivery note, or upload a PDF: the system recognises the document, extracts the data and checks it, just like in a real project. Free, no sign-up.

- URL: https://mojtabaamini.com/demo/data-extraction
- Language: en
- Author: Mojtaba Amini, AI Software Engineer (Verona, Italy)

## How this demo works

1. **Classification**: The model recognises the document type: invoice, contract, delivery note, order, quote.
2. **Extraction**: A language model (Llama 3.3 70B on Cloudflare Workers AI) extracts the fields and, for each one, copies the exact text it came from.
3. **Grounding check**: Every value is looked up in the original text: if the model wrote something that is not in the document, the field is flagged instead of accepted.
4. **Validation rules**: Deterministic checks: Italian VAT number check digit, plausible dates, line items that add up to the subtotal, subtotal plus VAT equal to the total.
5. **Export**: Verified data can be downloaded as CSV (for Excel) or JSON (for ERPs and APIs).

## Demo vs. a real project

- **Documents**: the demo reads text and digital PDFs. Real projects add OCR for scans and photos, and models such as Google Document AI.
- **Fields**: the demo uses a generic list. In a real project the fields and rules are those of your documents and processes.
- **Reconciliation**: in production the data is matched against the supplier, customer and product records already in your systems.
- **Integration and review**: data flows straight into your ERP, CRM or database, with a screen to confirm uncertain cases and a log of corrections.

Related service

- [Automatic document data extraction](https://mojtabaamini.com/services/document-data-extraction): From PDFs, invoices, contracts and scans to verified data in your ERP, CRM or Excel.

Product

- [Agent Flow Bind: 23 complete AI agents →](https://mojtabaamini.com/projects/agentflowbind-en): My agent platform for documents, phone and finance, with a live demo on agentflowbind.com.

## Frequently asked questions

**Is the data I enter stored?**

No. The site does not store the text: it is sent to an AI model on Cloudflare Workers AI for processing only. PDFs are read directly in your browser and are not uploaded. Still, avoid entering real personal data: use the samples to try it.

**Does it work with scanned PDFs?**

Not in the demo: it reads the text of digital PDFs. Real projects add OCR for scans and photos.

**Why are some fields flagged for review?**

Because they fail a check: the value is not in the text, the VAT number is invalid, or the totals do not add up. That is exactly the behaviour you want in production: flag a doubt rather than let an error through.

**Can I use it for my company?**

The demo shows the principle. Real use needs a pipeline adapted to your documents, checks and systems. Get in touch and we can talk about it.

**How many tries do I get?**

The demo is limited to a few requests per minute and has an overall daily limit. If you hit it, try again later.

## Let’s talk about your project

Tell me about the process you want to improve: I will reply with a first assessment of where AI can help and where to start.