A chat-first companion app that turns what you see through AI glasses into memories you can come back to
Overview
ROGI.AI is a pair of AI smart glasses with a companion app. I started with the PRD
I split it into what was broken, what the business needed, and what we only believed people wanted. That gave the team a shared start before anyone opened Figma.
Problem
People wear ROGI to keep a piece of their day without pulling out a phone
1. A growing pile of artifacts.Every session leaves photos, text, and voice. The app has to keep them and stay easy to navigate.
2. Privacy first.An always-on recorder breaks trust for the wearer and for the people around them. So ROGI captures only when the user presses the button, and a light shows when it's on.
3. Simple enough for anyone.Smart glasses easily feel like a gadget for geeks. ROGI had to feel as natural as sending a message, with no settings to learn.
4. When voice is not an option.In a meeting or on a loud street, the user has to run the glasses from the app.
Research
The product has to live through a whole day
ROGI is for work meetings, the small stuff in between, and time off. The main job: stay on top of the day without leaving the room and without looking covered in gadgets. If it feels unsafe or looks wrong on the face, nobody will wear it.
Jobs to be done
What job would the user hire ROGI for?
Not the glasses as an object, but the work they do in a life. I wrote one primary job, three jobs that keep it from breaking, and a filter for v1: if a feature does not help the job, it waits. The minimum was a photo from the glasses and a chat that knows that photo.
| Job | When | The user wants | In v1 |
|---|---|---|---|
| Stay present, without leaving the day or the look | In the flow of the day (meetings, deadlines, ideas, the city), something happens that cannot be lost | To capture it on the glasses, then find it or ask in the app, without looking covered in gadgets | Photo from the glasses + a chat that knows that photo |
| Trust the device and privacy | Wearing ROGI every day, and careful about data | To pair easily, see the status, and know capture happens only on the user's word | Battery, pair, consent, delete |
| Review the capture | Frames show up after a day of meetings and the city | To flip through them and come back to what mattered, not drown in Camera Roll | Grid, details, delete |
| Ask their own assistant | The user needs a cut or a next step from the day or the frames | An assistant that already has the user's context, so thinking starts faster than in a blank ChatGPT | Text + grounding on photos |
| Stay in the moment | The user sees something in the world and wants to understand it now | To ask through the glasses or the app without pulling out a phone, and live the moment instead of documenting it | A thin layer through chat + photo. Full skills later |
| Later, not the core of MVP | Skills in the moment, calendar, social, gift | Translate, what is this, a route. Tie capture to when. Friends. A helper for people close to the user | No. We do not build v1 around these |
Market research
I looked at what people already use for the same jobs
Pendants, camera glasses, note apps. Pendants turn the day into a transcript and a task list. Camera glasses want a post. Recorders listen when nobody asked. What was missing was quieter: keep only what the user chose, and help make sense of it later.
Personas
Four people, so we were not designing for “the user”
I sat with Lena in Berlin, Mark between cities, Sofya at school, and Irina in Copenhagen. We design for Lena. The other three catch what we miss: if Lena's Tuesday works, the same gesture should work for them.
Key findings
Six findings that decided the product
From the portraits and the jobs. Each one made it into the build.
1. Chat always in reach.The user can ask from any screen. The composer stays on the page.
2. Prompts follow the page.The model suggests prompts based on the screen the user is on.
3. Sessions are saved as the model decides.A glasses session is kept the way the model saved it, with its artifacts. We don't invent a special object for every case, just enough UI entities to hold whatever comes back.
4. A space that grows with use.Over time the app builds a personal space for each user from how they use it.
5. Privacy first.It is a real pain for the people who wear the glasses and for the people in the room.
6. A control panel when voice will not work.Voice doesn't always fit the setting, so the app needs a panel to run the glasses from the phone.
Flows and sketches
I mapped the paths, then sketched in code to try more than one idea
First launch, a session, the card, asking, shelves, the control panel. Static frames were too slow, so I vibe-coded rough sketches.
Usability testing
People used the sketches, and three things came back clearly
1. Ask from any screen.They wanted to ask without hunting for a chat tab, so the composer stays on every screen.
2. A place to handle the glasses.They wanted one place to manage the glasses, so the control panel got its own page.
3. Prompts from this session.They often didn't know what to ask, so the model suggests prompts from what the session kept.
Where we landed
Insight 1: Two surfaces
The app is two places: a control panel on the main screen to run the glasses, and a gallery of sessions that grows into a personal space.
Insight 2: Prompts from the open page
The model writes prompts from the open page and suggests the next action in chat.
Insight 3: Memory for recall anytime
Sessions from the glasses and from chat sit in a carousel by kind and grow into shelves. Over time it's the user's space, not a camera roll.
Considerations
How do we tell the user why a session started?
A lock screen notification could say what started the session, so the user knows why ROGI is recording without opening the app.
Next steps
Continued testing and implementation
The control panel, the session gallery, and the lock screen notification go into the next round of testing, and the high-fidelity screens go into the build.
Reflection
What I learned
The people in the room matter too.
We design for everyone around the wearer too. Capture only on the user's word and a light with one meaning are what let other people trust it.
Sketch in code, early.
Coded sketches let me try several ideas and drop one in a day, before it cost the team anything.
More work
