FinTech · AI-Assisted Due Diligence · 2025
I led design for a private-equity due-diligence platform where AI was producing analysis faster than analysts trusted it. As the design lead working with three ML engineers, I refused to ship the model without an "explain this score" surface. Analyst confidence in the AI's reasoning, not the AI's accuracy, was the bottleneck. Screening time dropped 60% (n=42 analysts, 90-day window) once the model showed its work. Client is under NDA. The methodology — making expert skepticism a feature rather than a failure mode — carries forward.
Want to know more about this work?
I can walk you through the decisions, the visuals I cannot host publicly, and the outcomes — under mutual NDA on a 30-min call.
Book a 30-min callDesign Patterns Used in This Case
This project crystallized two critical patterns for designing AI in expert domains:
- ML Explainability Patterns: Everything we learned about exposing reasoning chains—from reasoning exposition to peer-comparative analysis—became core patterns for making ML decisions transparent to non-technical experts.
- Human-in-Loop Patterns: The override-and-explain mechanism, risk factor weighting visibility, and respectful disagreement patterns are foundational examples of keeping expert humans in control.
One pattern per month with tradeoffs and code examples. 2,500+ designers already subscribed.