Meridian Line wants to move its multilingual crew-report retriever from its current embedding model to a different one that handles non-English text better. The Delta Sync index already holds two million chunks. What must happen for the switch to take effect correctly?
Select an answer to reveal the explanation.
Short Explanation
Each embedding model gives you its own map of meaning, so coordinates from different models do not line up. You’ll need to re-embed the whole corpus and rebuild the index; plan for that migration before promising the upgrade.
Full Explanation
An embedding model defines a vector space, and distance is meaningful only between vectors produced by the same model. To switch models correctly, you must re-embed the entire corpus and rebuild the index; with two million chunks, this is a significant cost and schedule item to plan for. Embedding queries with the new model while keeping old document vectors does not work just because dimensions match: two models can place the same sentence in entirely different positions, making cross-model rankings effectively random. Re-embedding only new and changed rows also leaves unchanged chunks with stale-model vectors, because Change Data Feed propagates source-table changes and generates no events for rows whose text has not changed. Swapping the endpoint’s model in place does not convert existing vectors on read; there is no automatic translation between embedding spaces. Exam caveat: treat the embedding model as a hard structural choice made when the index is created, not as a setting that can be changed without rebuilding. Practical check: evaluate candidate models on a representative sample of multilingual crew reports before committing the full corpus to one.