PRODUCTION SCENARIO
A Generative AI Engineer runs a parts assistant over an AI Search index. Questions that describe a part in words are answered well, but questions quoting an internal part code such as XR-4471-B come back with unrelated chunks.
Which change to the index query should the engineer try first?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Set query_type to hybrid so exact keyword matches score alongside embeddings
Hybrid matches the exact tokens in a query while still scoring vector similarity, which is what the docs point at when source data holds unique identifiers unsuited to similarity matching, and it keeps the description-based questions that already work. FULL_TEXT would find the part code but drops the semantic scoring those working questions depend on. Returning 100 results dilutes a list the identifier was never ranked highly in, and a larger embedding model finds little semantic signal in an opaque alphanumeric code.
Official Source: https://docs.databricks.com/aws/en/ai-search/query-ai-search ↗