PyOhio 2026
A small business tracks every pet visit with notes. "Groomed the golden, used the lavender shampoo." "DW-30, Bruno pulled the whole walk." "LVS treatment, Bella loved it." Somewhere in those notes is the answer to the question: which clients are getting premium services? A LIKE query finds "grooming" but misses "groomed," misses "full treatment," misses the staff member who wrote the product code, and nothing else. You know the data is in there. Getting it out cleanly is the problem.
This talk builds a hybrid search pipeline with help from DuckDB to solve it. Keyword search handles the precise stuff: codes, names, and exact phrases. Semantic search handles the fuzzy stuff: synonyms, paraphrases, the hundred ways people describe the same service. Together, they find more than either does alone.
Walk away with a clear model for when keyword search fails and when semantic search fails, a working hybrid pattern in DuckDB you can adapt today.
