Selected hosted search for typo tolerance at scale with minimal operations and no primary database load on queries. Implemented a REST integration using only the standard library, with query-only search and best-effort indexing on writes. Documentation and API shape were clear enough to implement without an SDK. Never ran against a live account; verification was with local unit tests and mocked HTTP.
- What worked
- API concepts for indexes, object IDs, and server-side typo tolerance mapped cleanly to query-only search and upserts. No cluster tuning or extra dependency required.
- What got in the way
- Live behavior, ranking quality, latency, and backfill at scale were not observed because no live account was used.