Concept guide · Embeddings and vector search
How do you choose an embedding model?
A slower explanation for when the short interview answer is not enough. Start with the intuition, then connect it to the technical idea and the way an interviewer expects you to reason about it.
ML area
Generative AI → Embedding model selection
Interview format
Foundations · Concept
Question
GENAI-034
Start with the idea
In simple terms
Choose an embedding model by testing whether it retrieves the right evidence from your own documents and questions. Dimension count alone does not tell you whether the model is good for your application.
Make it concrete
A support system with product codes, English and Hindi questions, and short FAQ answers may need a different model from a research system with long technical papers.
Build the answer
How to reason about it
- 1Define the retrieval cases and languages that matter in production.
- 2Compare quality together with latency, cost, dimensions, storage, and privacy.
- 3Version the model and plan re-embedding and rollback before switching.
Go one level deeper
The technical explanation
Compare models on a labelled retrieval set that includes languages, abbreviations, exact identifiers, long and short queries, and your real document types. Measure recall, ranking quality, latency, storage, cost, license, deployment location, and migration effort. Store the embedding model and preprocessing version with the index; changing either can require a new index and a controlled cutover.
Say this in the interview
Evaluate candidate models on representative queries, languages, document types, chunk sizes, and relevance labels. Compare retrieval quality, dimensions, latency, cost, licensing, privacy, and migration effort. Keep the model identity with every index because changing models requires compatible re-embedding.
Keep preparing