A lot happens between therapy sessions, but most of it gets lost before the next one.
I saw this up close with people I care about.
I wanted to help.
As a product designer, I could see what needed to be built, but shipping it alone was not realistic.
I decided to build it with AI, and this time, I wanted to involve AI in the entire process, not just the build.
The core challenge.
How do you help someone capture emotionally important moments fast enough that they actually do it, then turn those moments into something useful — without making it feel heavy, clinical, or invasive?
Involving AI did not replace the designing. It made it more efficient and expanded what a designer can do alone. That is not about cutting corners. It is about getting to users faster.
Standard Design Process
This Project
Ideation
With the help of ChatGPT, I combined research and ideation into one conversation. Within a day, I collected and sorted the pain points, pressure-testing feature ideas, and drafted a scoped PRD for MVP. Then I took that back to a real user asking for feedback.
Shipping Early
The first version of MVP was live within days, a real app, on real devices, in real hands via TestFlight. Testing prototypes can tell you whether a feature or the UX makes sense, but real use tells you whether it survives contact with behavior.
Scoping
The priority for MVP was clear: build for the people I care about, on the devices they use. A focused scope meant faster rollout, real feedback sooner, and room to expand to more devices and more users from there.
What Real Uses
Revealed
The Problem Was Not the Design
Some users stopped using the app for days because they found the recap summaries were not reliable. The on-device model was inconsistent for users to trust.
Better structure
Part of the fix was structure. By switching to specific output fields, the on-device model provide consistent recaps.
At the same time, users could scan exactly what they needed before a session instead of reading through a wall of text.
Key Worries, Turning Points, Open Loops, Questions for Therapist. Ready to discuss.
Optional cloud recaps
For users who wanted even better summaries, I added an optional cloud model via bring-your-own-token. Privacy stays intact, quality improves, and it remains entirely optional.
Cut the Feature Users Said They Wanted
Plans grouped recurring therapy homework under a named goal. Users called it a useful feature when they first use it, but they never did.
What usability testing can't tell
The first time users tried it, they liked it.
It felt useful and they said they would use it. What you get from testing is that when something is new, users are paying attention.
It cannot tell you what happens weeks or even days later, when the novelty is gone and real habits take over.
What real use showed
They stopped using it.
When I asked why, the answer was simple: repeating todos already covered the need. Plans added complexity that did not match how people actually thought about their week.
Simplifying the product
Removing Plans was not about the feature being wrong. It was about the product being cleaner without it. Sometimes the best product decision is knowing what to take away.
Insights That Used to Take Weeks
The original recap used calendar logic: daily recap, weekly recap. It made sense on paper.
What real use showed
People did not want "this week." They wanted the days since their last session. Last 3 days. Last 5. Last 7. Therapy does not run on calendar weeks. It runs on session cadence.
Summarize any range of days from 1 to 30 days, aligned to how your sessions actually work.
The difference
the new process makes
Traditional research could have found this eventually. Through interviews, synthesis, and iteration cycles, you would get there. But it would take weeks.
With a real product in real hands within days, this insight surfaced naturally through behavior. Users used it, and the pattern became obvious.
The process
used to stop at handoff
For a long time, the design process stopped at handoff. We designed, we documented, we handed off to developers, and then waited to see what got built.
AI expands
what designers can do
It gives designers the tools to stay in the process longer, getting real feedback sooner, and closing the gap between what is designed and what actually ships.
Real insights, much earlier
AI also enables a fully functional prototype at a very early stage of the design process. Which means we find insights much earlier, before too much has been built in the wrong direction.
The job is still problem solving.
The tools let us go further.
Try it Out
Capture the moment
Voice or text check-ins, so nothing important gets lost before the next session.
Skip the app entirely
Siri and App Intents let you log a thought without breaking your flow.
Recaps built for therapy
Rewind lets you review any range of days, aligned to how your sessions actually work.
Follow through
Todos with reminders keep homework alive between sessions.
Private by design
Local-first, cloud-optional, no account. Your data stays yours.
Therassist