Overview: Deal Analyzer turns a broker's deal package into a structured pre-screen. It extracts the data, checks it against the lender's rules, and drafts the questions to send back, so a commercial real estate lender reaches a go/no-go in minutes instead of the 24 to 48 hours their manual process takes. It is the product Hypha's leadership sees as most sellable, and the one the new sales team leads with.

Challenge: An early engineer-built skill proved the value but was not demo-ready, and more importantly it was not trustworthy to a skeptical underwriter. These are finance experts who will not take a number an AI pulled from a broker's self-interested email at face value. The design problem was to make an extraction tool both fast and believable enough for an expert to act on.

Solution: A pre-screen surface built around the expert's judgment: clean, traceable data first, with the AI's recommendation and buy box present but de-emphasized, provenance on every value, plus honest partial and empty states and a checklist that always shows what is still missing.

Impact: An end-to-end workflow, built and validated with the customer and handed to sales as the highest-confidence product to pitch. Broker response time compressed from roughly 24 hours to minutes. It also spun off a set of platform-level patterns (a generic Kanban, deal-type templates, email ingestion) that I am now driving cross-functionally across teams.