A housing authority indexes lease PDFs so a resident assistant can answer questions from those documents. Which retrieval and indexing approach should the team select?

A housing authority resident assistant that answers from lease PDFs must index that corpus for retrieval. Vector or hybrid retrieval over chunked lease content lets the assistant ground answers in actual lease language instead of inventing terms. For document-heavy municipal assistants, indexing and retrieval are planning prerequisites whenever the source of truth is a PDF library requiring extracted, chunked text rather than titles alone without clause content.

Indexing the lease corpus for vector retrieval with single-PDF embeddings to ground answers in document chunks fails because a single embedding for an entire lease blurs distinct clauses together and cannot retrieve the precise chunk needed for rent, deposit, or obligation questions. Indexing the lease corpus for BM25 keyword lexical search to ground answers in document chunks fails because exact-term matching misses resident phrasing, synonyms, and legal paraphrases that vector or hybrid retrieval over chunks is needed to capture. Indexing the lease corpus titles for vector or hybrid retrieval to ground answers in document chunks fails because titles alone provide no clause text to chunk and embed, so retrieval has no lease language to ground rent, deposit, or obligation answers in.

Exam caveat: hybrid retrieval may still need metadata filters (property, unit, effective date) so the wrong lease version is not grounded. Operational check: pick a known lease clause, ask it in resident language, verify the returned chunk cites the correct PDF section, and test that answers change when the indexed lease is updated.