Lawrean Kompa

How it works Extracts quotation text using PyMuPDF. Uses semantic matching to identify corresponding line items, even when descriptions differ. Classifies matches as confirmed, uncertain, or unresolved. Uses deterministic calculations to identify and verify numerical discrepancies. Provides supporting evidence to help users inspect the results.

Current limitations Maximum 3 pages per PDF. Maximum 10 line items per document. Maximum 10 MB per PDF. Text-based PDFs only; scanned and handwritten documents are not supported. One currency per document pair.

Who is it for? Procurement professionals, accountants, business owners, administrators, developers, and anyone who works with supplier quotations.

Current state

Kompa is an early-stage working prototype. It is available for public testing without registration.

I'm looking for honest feedback on matching accuracy, missed discrepancies, calculations, PDF extraction reliability, and the overall usability of the results.

Testing instructions – Lawrean Kompa

1. Access the application

Open the public demo:

No registration or login is required.

2. Prepare two quotation PDFs

Use two versions of the same supplier quotation:

• An original quotation.
• A revised quotation containing some changes.

Current document limits:

• Maximum 3 pages per PDF.
• Maximum 10 line items per document.
• Maximum 10 MB per PDF.
• Text-based PDFs only. Scanned or handwritten documents are not supported.
• One currency per document pair.

Please use sample or anonymised documents, not confidential business quotations.

3. Compare the quotations

Upload both documents and review the comparison results.

Pay particular attention to:

• Whether corresponding line items are matched correctly.
• Changes in item descriptions.
• Quantity increases or decreases.
• Unit price changes.
• Added or removed items.
• Incorrect, missed, or uncertain matches.
• Whether the calculated differences are correct.

4. Report your findings

Please share:

1. What worked as expected?
2. What failed or produced an incorrect result?
3. Were the explanations and supporting evidence clear?
4. Did you encounter any errors?
5. What would you improve?

Screenshots of unexpected results are especially helpful.

The goal is to challenge the accuracy and reliability of the comparison process. Honest negative feedback is just as valuable as positive feedback.