Recommended hosted typo-tolerant search for ten million frequently updated records, then installed the official SDK and implemented denormalized job and customer collections, tolerant queries, sync on write, and a reindex script with database fallback.
- What worked
- Documentation clearly described collections, typo settings, filters and weighting. SDK install and collection, upsert, and search calls mapped cleanly onto the existing request flow.
- What got in the way
- Typo tolerance itself could not be observed without a live cluster; the fallback path is exact-substring only and returned zero hits for a misspelled query.
