Service design, product strategy, and human-centered AI workflows
I interview people, trace the handoffs and business rules shaping their experience, then work out where a service or AI system should intervene and where a person still needs to decide.
Selected work
I lead with Lumen because it shows how I think when the problem is messy: make the evidence inspectable, define who is responsible for what, and be honest about what the work did not prove.
Lumen · Service model
Strategy teams needed a reviewable path from scattered research to partner decisions, not another untraceable AI answer.
Clinicians needed useful visual evidence at the moment of decision without adding workflow burden or transferring clinical responsibility to AI.
A surgeon-tested concept and pilot workflow were evaluated; no clinical efficacy, production deployment, or broad hospital adoption is claimed.
A distressed driver needed a clear path to help across web, hotline, WeChat, mobile, and in-car channels.
Final desktop and mobile designs and the modeled step reduction are documented; launch and production outcomes are not verified.
SMB customers needed a digital path through product, account, payment, subscription, and order rules previously supported by assisted sales.
Final designs and workflow decisions are documented; no checkout-specific production metric is sufficiently supported for publication.
One compact gateway had to help travelers choose among bundled packages and hotel-only stays without making the starting decision feel like work.
Both experiments moved targeted search behaviors in a positive direction. Exact percentages are withheld because the surviving records conflict on metric labels and values.
Sometimes it is repeated information. Sometimes it is uncertainty, a hidden business rule, or a judgment the technology should never make on its own. That distinction changes what should be designed.
Observe
Look at tasks, tools, evidence, handoffs, exceptions, and incentives.
Frame
Be clear about what technology may propose, what evidence stays visible, and who decides.
Adopt
Define ownership, rollout, recovery, guidance, governance, and measurement.
This project shows how I compare interaction models, work within an existing component system, and keep outcome claims inside the evidence.