Clean up a messy CSV without trusting a model blindly
A deterministic rules engine catches everything it can on its own; only the genuinely ambiguous cases reach an LLM. You review every proposed fix, grouped by pattern rather than one row at a time, and export a clean file plus a full audit log.
Drop a CSV here
No account, nothing stored publicly. Capped at 5,000 rows for this demo — a product decision, not a technical ceiling.
- Rules first
- Duplicates, formats, disguised nulls and outliers are resolved by pure functions — cheap, deterministic and testable.
- LLM only for the rest
- Near-duplicates and suspicious values go to the model with column context, never the whole dataset.
Want to see the review experience without uploading anything? Open the workspace on fixture data.