PRODUCTION SCENARIO
A Generative AI Engineer loads 1,150 maintenance manuals into a Knowledge Assistant. Answers about the 30 largest manuals, each 700 to 900 pages, never cite them, while answers about shorter manuals cite correctly.
What should the engineer change so those manuals can be cited?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Split each oversized manual into several files of a few hundred pages each
Knowledge Assistant skips PDF, DOC, DOCX, PPT and PPTX files with more than 500 pages, and files over 100 MB, during ingestion, so those manuals never entered the knowledge base at all and nothing in them can be cited. Splitting them under that limit is the only listed change that alters ingestion eligibility. Endpoint type changes scale and latency. Embedding context length governs chunks, not whole files. A Unity Catalog table with a file column is a supported knowledge source, but the same page limit applies to the files it points at.