AI-enabled origination · Work in progress
The work
I am currently leading design judgment and interpretation within an AI-enabled loan-origination platform project. Processors and underwriters work across dense documentation and domains such as credit, assets, and collateral. The intended capability is to help them understand extracted and compared information, reduce manual effort, and make better-informed decisions.
I am designing repeatable patterns and an internal design kit that can carry this information coherently across several agent domains. The work connects business needs, product intent, technical capabilities, brand standards, and UX reasoning. It pays particular attention to what a person can trust in an AI-generated view and where that person has agency to review or act.
The questions I want an interface to answer
What does the system know, and what is it inferring?
What is it suggesting, and what is it actually doing?
What does the person using it need to understand before acting?
How can that person review or recover from the result?
These questions define what I mean by a defensible interface. They are a design standard, rather than a description of the platform's implementation.
What defensible means
An interface decision should be explainable in terms of the business purpose, the product's intended behavior, the technology's actual capabilities, and the needs of the person using it.
That explanation should also make the tradeoff visible: what the decision serves, what it leaves unresolved, and what evidence could justify revisiting it.
Status
The work is ongoing and has not launched. The contribution described here is design judgment, interface design, and a repeatable pattern framework. Intended improvements in confidence, speed, or decision quality are goals, not measured results.
Project descriptions focus on my role and the contribution I can describe publicly. Confidential materials are omitted. Illustrations identified as reconstructions explain an idea; they are not original project deliverables.