An internal app built at Glean that ingests Gong sales-call transcripts, structures the contents into a spreadsheet, and summarizes the top requested features across calls. An early version conflated salesperson opinion with actual customer requests; the prompt was refined specifically to separate the two, so the output reflects verbatim customer asks rather than reps' own read on what customers want. A lightweight pattern for building a narrow internal pipeline over call-recording transcripts to surface aggregated customer feedback, iterated by prompt rather than by rebuilding the pipeline. See Long-Context LLM Ingestion for Unstructured Feedback Mining for the related technique of feeding large unstructured sources directly into a model, and Launch-Readiness Cross-Referencing Prompt for another Glean-internal AI app in the same family.