Pull every date out of an LLM answer and normalize it
Language models love to express dates in whatever style the surrounding prose suggests — “next Tuesday,” “06/04/2026,” “in a couple of weeks.” When you need those dates in a database or a calendar, you have to find them and convert them to a single canonical form. This tool scans LLM output, extracts every date and time expression, and normalizes each to ISO 8601.
How it works
The parser matches several families of expressions: absolute numeric dates, written-month dates, relative phrases (“today,” “in 3 days,” “two weeks ago”), weekday names, and clock times. Relative phrases are resolved against the reference date you set. Genuinely ambiguous numeric formats — where day and month could be swapped — are not guessed; instead the tool flags them and shows both possible interpretations, so you stay in control.
Tips and caveats
- Set the reference date to when the output was produced, not always today, or “tomorrow” will resolve incorrectly for older text.
- Trust the flags. A flagged
03/04/2026means the tool genuinely cannot tell month-first from day-first — decide from context. - Verify before automating. Heuristic extraction is excellent for triage but should be checked before feeding results into a scheduling system.
- Pair it with the number extractor to fact-check both the figures and the dates an LLM cites in the same answer.