Deep Prototyping
Working AI product experiences, real pipeline and real model calls, built fast enough that users can test it before engineering has spent a single sprint.
AI product design, built not deckware
AI product work fails on paper and succeeds in the browser.
I build the thing before engineering commits to it, so the direction gets tested while it's still cheap to change.
Book a 60-minute discovery call
No pitch deck. We look at your actual problem.
Principal designer, UX Director, ex-Google, ex-HubSpot (led product design for Breeze AI).
Hands-on, player-coach, 0-1 products and teams, from early-stage startup to large enterprise, consumer-to-internal tools.
What kind of help do you need?
Three ways inAI Tool Building
The working tool itself, architected from how an expert actually does the job, not from a generic spec of how the job is supposed to work.
AI Systems Teams Actually Use
The specific tools, workflows, and role-by-role changes built directly into how a team works, so an AI rollout doesn't stall as a pilot nobody adopted.
Specifically
A live AI assistant prototype running the real pipeline and real model calls, not a staged demo, that resolved a stalled design disagreement in weeks instead of a quarter and is steerable to a test phone in real time.
Specifically
Built the first working architecture for an enterprise AI tool from a domain expert's own workflow, voice memos and correction transcripts, ahead of the formal build phase.
Specifically
Audited a function's workflows role by role, then built the tools and role-specific playbooks that got AI actually adopted into sales and marketing work, rather than left as a plan for someone else to execute.
What changed because of it
Outcomes
01
A stalled disagreement over whether an AI assistant's feed would feel meaningful or read as noise is now in front of real users. The prototype, not a description of it, is what's generating their feedback and what stakeholders point to when asked if the direction is working.
02
A sales and marketing team whose work didn't fit any generic AI tool now runs on custom skills and tools built around their specific business, their people, and their technical constraints, incorporated directly into their workflows rather than sitting outside them.