A library catalog assistant finds topical books well but often misses exact ISBN lookups residents paste into chat. What retrieval approach should GenAIOps add?

Hybrid search that combines semantic topical retrieval with keyword matching for exact ISBNs is the approach GenAIOps should add when a library catalog assistant finds topical books well but misses exact ISBN lookups residents paste into chat. Semantic vectors excel at “mystery novels about cats” style queries; they routinely fail on dense identifiers where topical similarity is irrelevant and string match is decisive. Fusing keyword or BM25-style matching for ISBNs with vector retrieval for subject intent covers both resident behaviors in one catalog assistant.

Semantic vectors only fails because ISBNs are not topical concepts—embedding proximity rarely surfaces the exact 10- or 13-digit match residents typed. Disabling catalog search and telling residents to browse the OPAC abandons the assistant’s job and shifts load back to self-service without solving retrieval. A single embedding of the entire catalog with no per-title or identifier fields collapses granularity so neither topical titles nor exact ISBNs can be ranked as discrete documents.

Exam caveat: hybrid retrieval is the standard fix when stems contrast topical success with exact-identifier failure; do not confuse it with raising top-k alone or with replacing embeddings entirely. Civic catalogs, grant NOFOs, and ordinance section numbers share this pattern.

Operational check: after enabling hybrid fusion, measure exact-ISBN hit rate on pasted-identifier queries and topical nDCG on subject queries, tune fusion weights so keyword exact matches are not drowned by soft vector neighbors, and confirm OPAC metadata fields used for keyword indexing include normalized ISBN forms.