Handwave
I redesigned the mobile app and the palm-scan terminal so checkout and the app worked as one product.
Palm scan across hardware and mobile.
Product design
AI features people understand, trust, and keep using.
For products where the feature exists, or is about to, and people still will not use it.
Email me
Tell me what’s broken and where you’re stuck.
What you get
Designed so people use it
AI can automate a step. The design work is what makes a team understand what it can do, trust what it did, and keep it in the daily workflow. Most abandoned AI features fail on the first two.
Concepts tied to a metric
Journeys, prototypes, and success metrics in one thread: conversion, retention, support load, or time saved.
Ten years of product UX
Startups and enterprise. The craft is making complex work clear enough that people finish it.
Interfaces your team can ship
UI and a design system that stay consistent and accessible once engineering takes the work over.
Tested until the number moves
Five real users with the prototype before anyone calls it done, then the metric after launch. We refine until the number responds.
Designed for the moment the model is wrong
Every AI feature has a limit. The design decides what happens there: slow down, narrow the scope, ask, or say it does not know. Trust is won or lost at that moment, not in the happy path.
Examples of this work
Snapfast
I designed the brief so every recommendation stays attached to the posts that justify it.
The creator decides what ships.
Supliful
I built the design system and redesigned product setup so a new creator could launch a brand.
Supported $1.1M monthly GMV.
Casoo
I led the team that designed the casino from zero: registration, wallet, bonuses, and retention on web and mobile.
Designed for engagement and retention.
How it works
- 01
Name the metric
Conversion, retention, support load, or time saved. The design work is attached to one of those from the start.
- 02
Map the journey
Flows, states, and the moments where people drop off or stop trusting the system. For an AI feature, that means feeding it messy real inputs to see where the behavior breaks first.
- 03
Prototype and test
Put it in front of users. What they do beats what the deck said they would do.
- 04
Ship a system
UI and a design system your engineers can take over. Then we iterate until the metric moves.
The approach
Set expectations, then fail gracefully.
AI features lose people in two places: they never learn what it can do, or it fails once without saying so. The method comes from ten years of product UX and the AI product work in Lenny Rachitsky's archive, applied to your feature.
- Expectation
- The first screen shows what the feature can do with the user's own data, not a tour of the model. Amol Avasare, from Anthropic's growth team, calls routing people to the right feature with the right friction up front the single biggest lever in AI activation.
- Guardrails
- Feed the feature messy real inputs until it breaks, then decide what it does there: narrow the scope, ask, or say it does not know. One line in a system prompt, assign an owner only when someone confirmed it, removed the biggest trust issue in an AI summarizer. The fix is usually design, not a bigger model.
- Control
- Approvals only where the stakes are real. Ask people to click OK five times a minute and they approve everything, so confirmations go on the actions that cost money or reach customers.
- Evidence
- Five users with the prototype, then the metric after launch. Bob Baxley's line holds: many teams ship for years without watching a real person use their software. We watch.
Fit
This is for
Adoption is the problem
Startups and enterprise. You have a product, an AI feature, or a new flow, and people are not finishing it.
This is not for
A visual refresh
Not a fit if you only want a new look with no change to the journey, or if there is no one on your side to ship the work.
FAQ
Yes. The system is how your team ships after I leave. Supliful is the example: setup, the marketing site, and a system the company could grow with.
Usually one of two things. They never understood what it could do with their data, so they tried a generic prompt and got a generic answer. Or it was wrong once, confidently, and nobody designed what happens then. Both are design problems, and both are fixable without a new model.
We pick the number before the first screen. Then we test, ship, and watch that number. Pretty is not the success metric.
That is the default. I sit with product, engineering, and data. The work is not a slide deck from the sidelines.
Email me the product, the flow that is failing, and the number you want to move. I will tell you if this is a design engagement or if you still need discovery first.