Insurance Intake: Build an Answer Set from 100 Image Forms

The challenge

You've got 100 claim forms and a short summary, but no complete answer spreadsheet. Build a way to extract the claims and catch mistakes before accepting the results.

Start here

Open the SharePoint folder linked above. CLM-000001.pdf through CLM-000100.pdf are the inputs, also packaged in InsClaims_100.zip. insurance-claim-bulk.pdf is a three-page generator summary and preview, not a combined PDF of all 100 forms.

Open the first two individual forms and the summary. The first two individual forms are images with no selectable text. The summary contains selectable text and a preview of ten claims.

Try it

1. Define claim and expense schemas, then OCR the 100 forms without duplicating ZIP copies.

2. Validate claim IDs, amounts, dates, and expense totals; route uncertain fields for review.

3. Reconcile status counts with the generator summary and explain which totals it can actually verify.

4. Create a manually checked sample that tests the pipeline on different layouts and expense counts.

What to build

Build a claims CSV or JSON array, an expense table, an exception log, and a small annotated answer set. State how many documents and fields you checked manually.