A permit Q&A assistant must ground answers in the municipal code corpus. Which retrieval design best fits authoritative civic documents that need both topical and exact-match strength?
Select an answer to reveal the explanation.
Short Explanation
City code is both a library and a filing cabinet—you need the topic and the exact section number. Hybrid BM25 plus vector search gives you that mix for permit Q&A instead of guessing from memory or Twitter.
Full Explanation
Authoritative municipal-code Q&A is a grounding problem: answers must be retrieved from the city’s corpus, not guessed from parametric memory alone. A hybrid retrieval pipeline that combines BM25 keyword retrieval with vector semantic search best fits this setting because ordinance text mixes exact citations, defined terms, and paraphrased resident questions. Keyword channels catch numbers and titles; semantic channels catch intent when wording diverges.
Relying only on model pretraining without retrieving code sections risks outdated or invented provisions. Retrieving exclusively from social-media posts about city hall substitutes rumor for law. Disabling retrieval and answering from a hard-coded welcome message provides no ordinance grounding at all.
Exam caveat: hybrid retrieval improves candidate recall; the generator still must cite and stay faithful to retrieved chunks. Operational check: evaluate a set of exact-section and paraphrased zoning questions, confirm hybrid recall beats either channel alone, and verify answers quote or cite retrieved municipal-code passages rather than ungrounded model memory. Hybrid scores should be fused and deduplicated before generation so clerks are not shown conflicting snippets.