Back
App + System + UX Design

Matlab App for Behavioural and Physiological Experiments

  • Replaced a fragmented manual setup with a single system that runs a full hour of experiments unattended
  • Eliminated 7–8 researcher interruptions per session that were directly contaminating physiological measurements
  • Unified all data into one labelled dataset, making it easier to organise, analyse, and trace back
  • Halved study duration and cut expected participant dropouts by 80%

Study duration

−50%

4–5 months → 2 months

Dropouts

−80%

5 expected → 1 actual

Sessions

0

Interruptions per session

Data quality

Cleaner

Less disturbance, better signal

Role

Lead Researcher & Designer

Users

Participants · Researchers

Sample size

10 participants

Platform

Touch screen computer

Duration

8 weeks

Deliverables

MATLAB app · Conference presentation

The Challenge

Turn a short manual experiment into a one-hour automated study — without disturbing participants or affecting the physiological data.

Bright white buttons and misaligned layout were distracting — and affected pupil size as screens changed

Adding new SNR conditions meant stopping the session and making manual changes mid-experiment

Each manual switch introduced errors, strained the researcher, and disturbed participants' physiological state

One hour is at the limit of what participants can sustain — the design had to work with that, not against it

Methods

Revealing issues through observation, not just self-report

Participants rarely articulate confusion mid-task. Direct observation was combined with in-the-moment probing and post-session interviews to capture both behaviour and experience.

Observational Testing

Observed participants mid-session — watching for hesitation, pauses, and unexpected reactions that self-report would never surface

Progressive Piloting

Tested on myself first, then fellow researchers, then 10 participants in a pilot study before the full experiment

In-the-moment Probing

Asked in the moment when something looked off — "Is there anything you are looking for?" kept things conversational

Continuous Iteration

Refined continuously: changes after self-testing, then again after each participant round — no fixed iteration count

Synthesis

From observation to pattern

Field Notes

Kept written notes during each session — recording what participants stopped at, asked about, or visibly reacted to

Affinity Grouping

Grouped observations by type: unclear instructions, visual confusion, flow disruptions, emotional reactions

Key Insights

Interface clarity reduces cognitive load

When participants stopped to search the screen, the interface was competing with the task. Cognitive load belongs to the experiment, not the app.

Confusion rarely surfaced as words

Pauses and hesitations said it instead — each one needed a follow-up question. Behaviour reveals what self-report misses.

Researcher intervention contaminates data

Stepping in mid-session isn't just inconvenient — it shifts physiological state and contaminates the data.

Long sessions need designed pacing

At one hour, fatigue is expected. Breaks weren't a comfort feature — they were part of keeping the data clean.

Design Decisions

Every change came from observed behaviour.

Each entry names the observation that prompted it.

Experiment Flow

Before

Manual switching between conditions

Each new SNR condition required researcher intervention — up to 7–8 times per session. That's 7–8 disturbances, and 7–8 chances for error.

After

Automated progression across conditions

Continuous, standardised execution. Sessions ran without intervention for the full hour.

Break Timing

Before

Unclear waiting time

Participants didn't know when the next part would begin — adding uncertainty to an already long session.

After

Countdown + user-controlled continuation

Clear pacing. Participants knew where they were, and could continue when ready.

Break Content

Before

Blank screen or uncontrolled content

A blank screen creates discomfort; engaging content shifts emotional state before the next condition.

After

Neutral, narration-free video

Kept participants at ease without introducing new stimuli that could affect later measures.

Screen Design

Before

Bright white, high contrast

Affected pupil diameter as screens changed — directly interfering with physiological measurement.

After

Calm background, controlled light

Consistent visual environment throughout. Less stimulus noise in the data.

Interface Clarity

Before

Small targets, misaligned layout

Participants stopped mid-task to search the screen — adding cognitive load that belonged to the task, not the interface.

After

Larger elements, simplified layout

Reduced visual search time. Interaction faded into the background.

Impact

Faster studies. Higher retention. More reliable data.

Study duration

−50%
Typical: 4–5 months. This study: 2 months.

60 experiments completed in 2 months. The same volume typically takes 4–5 months.

Dropouts

−80%
Expected: 5. Actual: 1.

1 dropout from 30 participants. Based on prior experiments, we budgeted for 5.

Sessions

Uninterrupted hour-long sessions

No researcher intervention required from start to finish.

Workload

No manual adjustments

Eliminated 7–8 manual condition switches per session, and all associated error risk.

Data quality

Cleaner physiological data

Less emotional and physical disturbance during measurement improved signal integrity.

Reflection

What I'd do differently

  • Being both researcher and designer on the same study meant every decision had to serve data integrity and participant experience at once — more demanding than either role alone.

  • I'd document each change more systematically next time — noting what changed, why, and what it replaced. UX documentation was secondary to building a system that served the experiment, and it shows.

  • Participants varied widely. Some were cooperative to the point of bias; others disengaged. The design had to hold up regardless.