Normalises messy records, finds duplicates and explains every change.
Data Cleaner takes a batch of records and returns them standardised: names cased properly, addresses normalised, phone numbers in E.164, companies deduplicated across spelling variants, and a change log explaining every edit it made.
The output includes a diff. For each field the agent changed, you get the original value, the new value and a one-line reason. Data cleaning that you cannot audit is data corruption you have not noticed yet.
Deduplication is where naive tools fail hardest. String similarity says "Acme Corp" and "Acme Corporation" are different, and that "Acme Ltd" and "Acme Limited" are the same as each other but not as the first two. The agent reasons about them as company names, and it knows that "Apple Inc" and "Apple Records" are genuinely different organisations.
It returns a proposed cleaned set alongside the original. Applying the changes is your step, not its. For a first run against a production database, apply the diff to a copy and read it.