Synthetic Datasets

Tackle blurry text, a broken archive entry, and fields that disappear in a scan.

Start batch PDF extraction with short forms and CSV and JSON answers to check your work.

Practice OCR, keep leading zeroes in ZIP codes, and learn when a field simply is not on the page.

Build a batch job that can pick up where it left off, then check tracking numbers and addresses.

The PDFs cover only part of the 14,956-transaction dataset. Finding that gap is part of the challenge.

Start with a receipt PDF, build a sales ledger, and uncover a pattern in loyalty discounts.

Practice handling large files, unpacking nested items, and checking what the data actually covers.

Practice finding several receipts on one page and checking their item totals.

Six overlapping exports turn into one pull history and a chance to explore probability.

Two PDFs and one model answer introduce field mapping, nested expenses, and simple checks.

Practice OCR when a complete answer spreadsheet is not available.

Use 100,000 sessions to study pageviews, visit times, and repeated trips to the Blog page.

Explore two ways to recover 15,347 receipts and fix a CSV with broken quoting.

A small import exercise where the file extension does not always tell you the format.

Build a compact graph from 47,002 transactions and keep user labels separate from your review rules.

Practice table extraction with shuffled text, gaps in IDs, and amounts in several currencies.

Catch duplicate files before they double your totals or sneak into both sides of a model evaluation.

A smaller payment network with tricky number formatting and useful lessons about risk labels.

Work through 2,392,287 transactions without mixing up store IDs or counting receipts from preview PDFs.