Work / SereniBrain

From session viewer to practice coach

The hardware is finished. The app is the opportunity.

Product Design· Research· UI· Self-initiated · 2026

SereniBrain ships a brain-sensing meditation wearable with a companion app that records everything and reveals almost nothing. This concept redesign turns the data the device already captures into the product's real value, and its strongest, lowest-cost growth lever.

Before and after of the SereniBrain home screen. Before: a session viewer with a brain-computer interface explainer carousel, a meditation tracker, and group meditation. After: a Settle Forecast that says it is a strong window to sit, with streak, typical settle time, average deep calm, and a when-you-go-deep weekly heatmap.
SereniBrain home · before and after

01  The problem

An insight-poor app on great hardware.

SereniBrain's engineering gravity sits where you'd expect for a neurotech startup: the EEG headband. Dry electrodes, signal processing, brainwave classification. That is the moat, and it is genuinely good.

The companion app, by contrast, behaves like a data read-out for the hardware rather than a product in its own right.

What the app shows todayWhat a daily practitioner wants
One score per sit (33 · Poor) and a Calm / Relaxed / Active donut
Am I getting better, settling faster, over weeks?
A live state curve, shown once, then forgotten
When do I practice best, and why?
Past sessions listed as durations, no trend, no comparison
What's my minimum effective dose?
An Export button as the only path to the bigger picture
Is today a good day to push for depth?

Every one of those questions spans many sessions. The app only ever shows one.

A spreadsheet of exported SereniBrain session data, with columns for calm, relaxed, and active state percentages and brainwave bands across dozens of sessions
To answer those questions myself, I exported four spreadsheets and roughly 400 MB of raw samples (brainwave voltages logged every 80 ms) and wrote my own analysis. The data was rich, longitudinal, and beautifully sampled. It was simply trapped one layer below the surface.

02  Context of use

I designed for the moment, not the metric.

I grounded the redesign in three months of continuous practice (59 device-recorded sits plus a daily reflection journal) and, just as importantly, in the situation those sits happen in. Focusing the context of use, not just the numbers, reframed the whole product.

Average calm by day of week, reconstructed from the export

30
M
54
T
45
W
25
T
58
F
45
S
37
S

Thursdays are tough. Depth varies sharply by day, a pattern that is invisible in a session-at-a-time view.

03  The reframe

Answers to context-of-use findings .

Before the redesign the app is a session viewer: it grades the last sit and waits to be exported. After it's a practice coach that remembers, compares, and tells you what to do next, in your own context.

01

Surface what you already capture

No new sensors. Compute settle-time, trends, and best-window from data already logged. The lowest-lift, highest-value move in the product.

02

Compare me to my own baseline

"52% calm" is noise. "Above your 42% average" is feedback. Every number is framed against the user's own history.

03

One number that matters

Lead with time-to-settle, the metric tied to the user's actual goal (deep absorption, faster), not an opaque 0 to 100 score.

04

Capture context at the moment of use

A two-tap pre-sit check-in closes the loop the journal was already trying to close, and makes every later insight sharper.

04  The redesign

Four screens, built around the moment.

Insights · the cross-session view

The details users really want

A time-to-settle trend, a best-window heatmap, and plain-language cards: arriving alert is your number-one driver, Thursdays are tough, your sweet spot is about 20 minutes.

Session report · from grade to feedback

A grade becomes a conversation

The old screen delivered a bare "33 · Poor" and a donut. The redesign opens with a plain-language headline, then three numbers that each compare to the user's baseline, and ends with a one-line reflection prompt.

Before
The old SereniBrain session report: a Performance Score of 33 labelled Poor in a donut with calm, relaxed, and active percentages, and a practice state curve.
After
The redesigned SereniBrain session report: a plain-language headline reading one of your deepest sits, three numbers compared to baseline, a session flow curve, and a brain-state band.

Pre-sit check-in · the one new input

Two taps for the highest ROI

Before starting: how are you arriving (alert, neutral, sleepy) and which practice. This is the only added data capture in the whole redesign, and the most valuable, because arrival state predicted depth better than any sensor signal.

05  The business case

The cheapest growth lever they're not pulling.

Hardware is a one-time sale with thin margins and a returns risk. The app is where a neurotech brand earns recurring revenue, retention, and word of mouth. SereniBrain is under-investing in exactly the layer that compounds. The redesign is attractive precisely because the hard part is already done.

A premium "Insights+" tier

Single-session feedback stays free; trends, best-window, forecasts, weekly digests, and unlimited history become a low-cost subscription. The asset already exists in the export, it just isn't packaged or priced.

Retention and engagement

A coach that gives you a reason to open the app before every sit, plus a streak and forecast to protect, lifts session frequency and cuts the churn wearables face after the novelty fades.

Shareable insight as free acquisition

"My brain settles 40% faster on Fridays" is screenshot-worthy in a way a donut is not. Personal-progress cards turn users into a marketing channel for the hardware.

Higher perceived hardware value

When the band visibly makes you better over time, the purchase feels justified. Insight reduces buyer's remorse, protecting the core hardware revenue directly.

Every feature here is computed from data already collected. No new hardware, no new sensors, no ML retraining. The lift is design surface area, not R&D, which puts the whole redesign in the rare low-effort, high-impact quadrant.

Note: the revenue and retention framing above is stated as opportunity hypotheses to be validated, not measured results from a shipped product.

06  The hard parts

Digging for insights.

The redesign looks tidy in hindsight. However, most of the actual work happened upstream, in the unglamorous business of getting the data into a usable state and then figuring out what to ask of it.

01

Getting the data out, and getting it clean

The single hardest part of the project was consolidating a long trail of separate files by hand. My journal was worse: transcribing months of handwritten entries that existed only on paper for essential arrival state context data.

My moveI did the manual work, transcribing the journal, merging everything into one place, and deciding entry by entry what was signal and what could be omitted. Then I checked for duplicates and entry errors.

02

Knowing which questions to ask, and what to ignore

Clean data is not the same as insight. I knew ideas weren't going to leap out of those numbers from staring at them. I had to cut out the filler and decide what to disregard.

My moveI went back to why I meditate in the first place. Luckily I record annual goals for each area of my life, and could reference them to build a list of what actually mattered. I then framed the analysis around these questions: does consistency matter more than session length, do longer sits create deeper calm or just more fluctuation, am I settling faster than I was in February, what is my minimum effective dose, and what conditions surrounded my best sessions?

03

Prompting for augmentation, not autopilot

Asking a model to "analyze this data" hands it the job of deciding what matters. It will surface what it considers important, which is a fine starting point on some projects but it was the wrong one here, where the whole point was meeting a specific personal goal.

My moveBefore using AI on anything, I define the goal and set the parameters so it enhances my analysis instead of replacing it. I fed it my actual questions and the context behind them, asked it to surface recurring themes and trends, and required it to cite where each conclusion came from so I could check if anything sounded off. An AI output is only as good as the precision you bring, which is why I save this kind of work for when I have the mental energy to be exact.

04

Turning trends into a product

Then came the moment of truth: now that I had the patterns, what do I do with them? The first dashboards the analysis produced were not good, which was clarifying. It told me this needed to be a designed, visual thing rather than a report.

My moveI started where all my designs start, with a rough pencil sketch in my planner. Then I got specific about the ask, requesting practical improvements that would be a "low architecture lift" and would help the user identify patterns, reflections, and actionable insights. I pointed to an app I admired for reference, and fed in my sketches alongside what the current app actually looks like. The more context I gave, the better the output came back.

05

Making the case beyond the mockup

A mockup shows what could be built. It does not explain why a hardware company should spend anything on it: the argument that decides whether a redesign ever happens.

My moveI pushed past the interface and into the business case: the hardware here is genuinely good, but the software is where a neurotech brand earns recurring revenue, retention, and word of mouth. Framing what the company is leaving on the table by under-investing in the layer that compounds is what turns a concept into a proposal.

Every one of these was a judgment call rather than a technical step, and each one shaped the product more than the visual design did.

06  Reflection

What context of use taught me.

The most useful move in this project wasn't analysing brainwaves. It was asking who is reading this, when, and in what state. The same dataset that produces an intimidating 400 MB export becomes a calm, two-line morning nudge once you design for the moment instead of the metric.

If I took this further

  • Usability test the groggy-morning home with real practitioners: is the forecast trusted or ignored?
  • Validate the forecast model against held-out sessions before promising it on the home screen.
  • Instrument and A/B the pre-sit check-in: does context capture actually improve perceived insight?
  • Price-test Insights+ and measure its effect on 90-day retention.

What I'd carry forward

  • Lead with the one metric tied to the user's goal, not the one the device finds easiest to output.
  • Frame every number against the user's own baseline.
  • Treat recurring software value as part of a hardware product's business model, not an afterthought.
  • Design the moment of use, then let the data serve it.

Good data is necessary. Knowing the situation it lands in is what makes it useful.

Concept · self-initiated · 2026

See the rest of my work.

View all work