Toronto
Product designer at Stellantis, leading advanced speed-control features for ADAS and HMI. I work where attention is scarce: in the car, the inbox and the headset.
0km/h
Your scroll speed, limit 120
Scroll fast and watch the sign.
Four projects, one question: what deserves attention right now, and what can wait?
Stellantis, in production
Cruise control that adopts the legal speed limit, and a driver who stays in charge.
What gets throughThe new limit, only when the driver wants it.
Stellantis, 2026
An AI tool that lets HMI teams decide on evidence, for any feature. The overspeed warning was its pilot.
What gets throughEvidence, not seniority.
Evaluation report
Overspeed warning, full-size SUV cluster
Driver panel, 30 drivers
22 keep it on8 neutral0 disable
University of Toronto, VPRI, 2022
A 9-year-old funding alert service, redesigned around where people read it: the inbox.
What gets throughThe alert that matters, not five a day.
University of Toronto, 2021
A focus app for mixed reality glasses: a quiet office you can wear.
What gets throughThe call you’re waiting for, not the room.
Control earns trust.
Drivers keep a manual mode and a way back to their own speed, because no speed-limit data is perfect.
From Intelligent Cruise ControlEvidence before opinion.
Every score tied to a source, so a decision stays decided.
From CoDesignerLet research set the scope.
The brief said website. Interviews said inbox. We designed the inbox first.
From Research AlertsDesign the boundary, not just the setup.
What matters is the moment the real world gets back in.
From Aug-FiceHello, I’m Sahil.
A user-centred product designer with a decade across engineering and design, in startups and global enterprises: mobile, web, automotive and gaming.
Vision
Experiences that feel almost invisible, yet profoundly impactful.
As a photographer I see the world through different lenses, and people’s stories keep reshaping how I see it. The Himalayas taught me tranquillity, sustainability, simplicity and adventure.
Mission
Co-design with communities, at every step.
Empathy, accessibility and clarity in every interaction. Inclusive designs that inspire confidence, encourage meaningful features and reflect the diversity of human stories.
Design, digital marketing and quality assurance
Quarterly global UX meet, Stellantis. For a working file that settled, in one meeting, logic that used to take weeks.
Design roles, HMI, research tools, or anything in between.
ADAS HMI case study, Stellantis
Speed Limit Assistance lets Adaptive Cruise Control and the Speed Limiter adopt the legal limit, with the driver still in charge.
Drivers pay for speed-limit features, then switch them off.
Benchmarks, social listening, journey maps, a guerrilla test and a 30-driver study.
Launched with the driver in control. Next: retire the popups.
Speed signs pass. Accept a new limit with RES, go back to your own speed, or send a critical alert and see who gets the screen.
An illustration of the behaviour, not the production HMI. The planned version is a direction, not the final design.
Existing solutions were oversimplified and poorly received.
Drivers still struggle with cruise control and speed limiter basics.
A new safety function has to be obvious and easy to switch on.
Speed-limit data is unreliable across the industry.
Drivers often go over the limit on purpose on highways.
A multi-team pipeline meant launch risks and inconsistencies.
Eight stops, and it doesn’t end at launch.
PlannedNext loop: retire the popups
1
The factory setting keeps the driver in control. The feature is visible, but it never forces a limit.
2
No speed-sign recognition is 100% accurate, and you’d likely spot an error on your daily commute. So the driver can resume their previous set speed.
90%+ of the people in the automotive forum threads I analyzed had turned off a feature they paid thousands for.
Nobody else in the industry offered this. The team was afraid of being on an island, so I made the case with evidence: forum research and two journey maps.
Journey A
The car reads the wrong speed limit.
Journey B
The driver wants to keep up with traffic, above the legal limit.
We chose discoverability
123
A 3-step activation, more complex on purpose.
Drivers chose simplicity
12
2 steps, even with a popup at every speed-limit change. We shipped it.
As ADAS continuous-improvement lead: what happens when a driver wants to leave the legal limit?
Through the continuous-improvement loop I had built, I got new requirements approved.
Planned
That was the clearest message from the latest round of user feedback. The popup space is shared with messages that matter, so it should stay reserved for critical alerts, not feature HMI like this.
Launched
A popup for every new limit
Next
A subtle animation in the sign itself
Built with the UX engineering team’s hi-fi prototypes. Far less ambiguity, far fewer development issues.
They had never used them. I showed how they win an argument.
Not just for hand-off. I helped the team learn it for every stage of development.
“Silent hero” Quarterly global UX meet
My working file for logic that used to take weeks of meetings. It was settled in one.
There is more to do as next-generation ADAS modules move from L2+ to L3 and beyond.
ADAS HMI case study, Stellantis, 2026
An AI decision tool for HMI teams. Any feature, any design question, settled on evidence instead of the loudest voice in the room.
ExamplePilot feature: overspeed warning
138
km/h
18 km/h over the limit
I designed CoDesigner end to end, piloted it on 1 real decision, tested it with 6 designers, got 2 directors on board and walked 60 people through it at a design review.
14 brands, 14 playbooks. When evidence was thin, the meeting decided, and decisions came back months later.
A 14-step design thinking loop, then a working prototype built with Claude from my specs.
A study that ranks designs for any ADAS feature and shows why. The pilot replayed the overspeed warning. Leadership backed it.
14 brands, each with its own way to choose HMI designs. When evidence was thin, the meeting decided. Decisions came back months later: launch delays, millions lost.
Today
each team argues its case
seniority settles it
reopened at the next gate
What the team wanted
all in one study
recommends, with reasons
Speed Limit Assistance, a feature I led. Eleven stops, and it still isn’t done.
How it went
The goal, with CoDesigner
Same study, any ADAS feature or HMI question. Pick one.
Ranked, with reasons. Bars are an illustration; only the overspeed pilot ran for real.
Fourteen steps, from a first chat with my manager to a review with 60 people. Red explores, teal builds. Select a step.
Start
Now: building the data foundation
1Project rundown
In a 1:1, I asked my manager why we didn’t use a more structured approach instead of going into reviews unprepared, and whether there were heuristics or best practices we should follow. He put it down to the merger, and I could see why.
Problem
No shared, evidence-based way to choose HMI designs across regions.
Who it’s for
UX designers who lead one feature for its whole life, and the directors who sign off.
Constraints
No engineering team or budget yet. Prove the value with a working prototype first.
Success looks like
Decisions the team can defend, reuse across brands, and stop relitigating.
Pilot
One live ADAS decision: how the cluster warns a driver who is over the speed limit.
2User research and personas
Here, the HMI feature designer owns a feature across brands, architectures, regions and model years. Every call they make follows them into every brand.
Proto-persona
Maya
ADAS HMI feature owner for the speed control domain, across Stellantis brands.
“I own this feature in every brand we sell. I need decisions I can defend in any room.”
Systems engineering, homologation, product planning, quality, benchmarking
Global UX director, ADAS director
| Tool category | Finds internal knowledge | Customer signal | Competitor benchmarks | Scores HMI heuristics | Checks regulation and architecture | Explains the decision |
|---|---|---|---|---|---|---|
| Enterprise AI assistantsCopilot, ChatGPT Enterprise | Partly | No | No | No | No | Partly |
| Knowledge platformsNotion AI, Atlassian Intelligence | Yes | No | No | No | No | No |
| Customer insight toolsQualtrics, Medallia | No | Yes | No | No | No | No |
| Benchmark providersTeardown and feature databases | No | No | Yes | No | No | No |
| Synthetic research platformsSimulated user feedback | No | Partly | No | No | No | No |
| CoDesigner | Yes | Yes | Yes | Yes | Yes | Yes |
covers itpartlynot covered Existing tools each cover a slice. None scores HMI designs against heuristics, regulations and architecture together.
3Empathy map
Pains Opinion beats evidence. Sources scattered across teams. No shared scorecard.
Gains Confidence to push back with data. One design reused across brands. Time back for design.
4Workshop and playback
A workshop with the global HMI team: pain points, clusters, playback. One cluster ran through all the others.
The loudest voice wins
Seniority settles ties
Nobody can say why we picked it
Reopened at every gate
No time to read papers
Clinics cost $100k+
Heuristic reviews get skipped
Benchmarks live in personal folders
Owner’s manuals hunted one by one
Quality data is hard to get
Architecture limits found after design
Regulation checks come last
Every brand has its own guidelines
Same study run twice in two regions
No scorecard for variants
Heuristics exist, nobody applies them
We recycle last program’s design
Few outside perspectives
Decided by committee, not by data or heuristics. It shows up in every other cluster.
5As-is scenario
Main user: Maya, feature owner. Her expectations: ship one design that holds up in every brand, decide on evidence, reuse what other regions learned.
Phases
Kick off
Gather evidence
Create variants
Compare
Align
Build and learn
Does
Feels and thinks
Pain points
Every brand and region brings its own playbook
Evidence is scattered, so research gets skipped
Few perspectives; last program’s design gets recycled
No shared scorecard, and clinics cost $100k+
Main pain point Opinion and seniority decide, not data or heuristics
Late surprises; learnings aren’t reused
Opportunity
One starting point per feature, with rules by brand and region
Sources gathered and ranked by credibility
Real benchmarks plus AI-generated options
Score variants on heuristics; early driver signal
A traceable rationale that keeps decisions closed
Check constraints up front; log every decision
6Needs statements
Decided by committee
A feature owner needs a way to compare variants against shared heuristics so that evidence outranks seniority.
Evidence is scattered
A feature owner needs a way to find benchmarks, regulations and past decisions in one place so that nobody starts from zero.
Late surprises
A feature owner needs a way to see architecture and regulation limits up front so that late surprises stop forcing rework.
No time for research
A feature owner needs a way to get directional driver feedback early so that only finalists need a $100k+ clinic.
Many playbooks
A feature owner needs a way to share a rationale anyone can trace so that a decision stays closed and every brand can reuse it.
7Big ideas
We pushed past the obvious on purpose. Even the wild ideas pointed the same way: the team wanted evidence that carries weight in the room.
Searchable benchmark library
Best-of-the-best decision log
Brand and region rule packs
Architecture constraint lookup
HMI heuristic scorecard
Synthetic driver panel
Evidence-strength meter
Explainable recommendation report
Blind review: hide who designed what
AI-generated concept options
Chat assistant: ask the HMI brain
Read designs straight from screenshots
Decision roulette for alignment meetings
An AI clone of the ADAS director that pre-approves designs
Opinion tax: $5 every time someone says “I feel”
A/B test new warnings on the whole fleet tonight
One global HMI czar decides everything
A human-factors professor for every brand
made the first versionwild card
8Prioritization
Ideas plotted by impact on decision quality, and by whether a prototype could prove them now.
Needs a data foundation
Build first
Park
Cheap but shallow
Higher impact
Lower impact
Harder to prove now
Easier to prove now
Later: the real repositories behind the library, regulation checks, reading designs straight from screenshots.
We ideated from sensible to absurd (an AI clone of the ADAS director, an opinion tax of $5 every time someone says “I feel”), then plotted every idea by its impact on decision quality and whether a prototype could prove it now. Five made the first version.
9To-be scenario
Main user: Maya, feature owner. What she gets: one study instead of scattered research, a score for every criterion with its source, and a decision that stays decided.
Phases
Kick off
Gather evidence
Create variants
Evaluate
Align
Learn
Does
Feels and thinks
What changes
No more starting from zero
Evidence before opinion
A wider option space
Every score is explainable
Decided once
Knowledge compounds
Risks to manage
Rules must be kept current
The repositories have to exist first
AI options need clear labels
Trust in synthetic panels
Leaning on one number
Back-test before scaling
10Hills
11Storyboard
Six frames, told with the pilot feature. Swap in any other.
1A new speed-assist rule lands.
2Maya opens CoDesigner and picks the feature.
3Her variants, plus real benchmarks and AI options.
4She builds a driver panel from real segments.
5Sign blink wins, refined to gray-to-black.
6One meeting. Decided, logged, reused.
12Prototyping
Rough sketches of the flow went back to the team before anything was built. They confirmed it was what they needed, so I moved straight to a working prototype.
new study
driver panel
results
how we judged
ask CoDesigner
why did A win?
answer + [K1] [S07]
ask anything…
Early sketches, redrawn.
I wrote the specs, flows and scoring rules, and Claude wrote the code. The pilot replays a decision the team had already made, so reviewers could judge the output against a known answer.
v1
Built
v2
Built
v3
In design now
| Source | What we heard | What changed |
|---|---|---|
| Leadership | Use our own HMI heuristics, and put a number on every decision. | Org heuristics feed every 1 to 5 score; weights roll the scores up to one total. |
| Team | AI options, or real competitor benchmarks? | Both, side by side and labelled. Some AI options were worth testing; the designer decides. |
| Pushback | Synthetic users aren’t real users. | Panels come from real segments, with minimum sizes, and every answer can be traced. |
| Usability | A complex decision tool is a lot to take in the first time. | A chat assistant and smart defaults, in design now. |
| My learning | AI must never invent evidence. | Only cited, approved sources count, with weights the team controls. |
Inside the prototype
Click through the study the way Maya would.
Overspeed warning, full-size SUV cluster. 30 drivers, 34 research sources, 16 org knowledge entries.
Recommended design
Speed-sign digit blink Regulation: compliant
Digits blink inside the speed-limit sign drivers already glance at: noticed fast, low effort, never competes with collision or lane alerts. 22 of 30 drivers would keep it on.
Red digit blinkBaseline
Gray-to-black digit blinkRecommended
Red outline blinkCompetitor reference
22keep on
8neutral
0disable
“It uses the sign I’m already glancing at. Subtle but unmistakable.” Persona W20, 54, comfort-focused owner
Overspeed warning, full-size SUV (NA). Driver panel: all owners (30).
Speed-sign digit blinkBenchmark, 5 OEMs
Pulsing red numeralBenchmark, 1 OEM
Gradient warning barAI option
Ambient peripheral glowAI option
Your designs, real competitor benchmarks and AI options in one set, each labelled by source.
All owners, 30 participants. Ready for a go or no-go call.
Participants30Ready for go / no-go
Average age47Range 29 to 71
Tech comfort9 / 11 / 10High / moderate / low
Alert tolerance11 / 14 / 5Low / medium / high
Trim mix16 / 14Top / core trims
Regions23US states
Panels start from segments grounded in owner data. Under 8 people is flagged as directional only.
31 of 34 included, 3 excluded.
Glance timeStrong11 sources
Cognitive loadStrong5 sources
Response timeStrong6 sources
AdaptabilityStrong8 sources
AcceptanceStrong14 sources
Makes speed assistance mandatory; the warning must be perceptible without being unduly distracting.
High credibilityTasks that compete for the same channel interfere most; ambient cues load the driving channel less.
High credibilityAdvisory cues must never cover or compete with safety-critical alerts.
High credibilityAnecdotal reports of drivers switching off nuisance alerts.
Low, excludedRegulations, standards, papers, surveys, forums and org knowledge, each weighted by credibility. One click drops the forums.
The shared knowledge every evaluation draws on. Admins curate it; study teams can still add research for one test.
Entries18Across 7 categories
Applied16Feeding every evaluation
Drafts2Awaiting approval
Last updateSep 252026
Architecture rules, brand guidance, heuristics, past studies, best-of-the-best decisions and benchmarks, curated by admins.
A chat assistant docked beside the report. It explains the scores in plain language, cites its sources, and turns follow-up questions into new studies.
Try a question
Answers cite approved sources only.
Recreated screens, program anonymized. Numbers are the prototype’s own demo data.
13Evaluate
From the team that asked for it, to the people who would use it daily, to the directors who would fund it.
6 ADAS designers and feature leads, North America
Global UX director and ADAS director
Both saw it making the team’s decisions more confident, and giving designers backing to push for their ideas.
A fair challenge. The answer was to show where the panel comes from, and where it stops.
In-house clinic
CoDesigner panel
The fastest honest check: replay past decisions with known outcomes and see how often the tool agrees.
NextThe pilot already replays the overspeed digit-blink decision end to end. Running more past decisions through it is the next validation step.
14Playback theatre
I presented the problem, the process and a live run of the prototype to 60 people across different domains.
60 people
Leadership’s read: more confident decisions, and backing for designers to push their ideas.Global UX director and ADAS director, paraphrased
The AI is only as good as the data it reads. Building CoDesigner forces us to streamline how benchmarks, compliance documents and decisions flow.
Scattered today
Benchmarksscreenshots in personal folders
Compliance documentsmany versions, many owners
Best-of-the-best patternsburied in slide decks
Alignment decisionsmeeting notes, if anywhere
Curated repositories
Benchmark libraryevery teardown, tagged
Regulations and compliancecurrent version, one owner
Best-of-the-best libraryapproved patterns with rationale
Decision logevery decision and its evidence
CoDesigner
Decision with rationale
It should only stand on reliable, cited sources, with weights the team can adjust.
Real benchmarks and AI options side by side, and the designer decides what counts.
The repositories are the real product. The interface is the easy part.
Chat and smart defaults on day one, full control by day ninety.
$100k+The cost of one in-house clinic. Narrowing concepts before a clinic is where the savings start, and over time they are projected to reach millions.
All artifacts are redrawn and anonymized for confidentiality. Product screens recreate a vision prototype and use its demo data.
UX research and service redesign, University of Toronto, 2022
Redesigning a university’s research funding alerts around the place people actually read them: the inbox.
Final prototype, one screen per redesigned stage. Select a stage, or open any screen full size.
A team of six spoke to 9 people on both sides of a 9-year-old service, mapped 5 journey phases, redesigned 3 touchpoints and tested them in 2 task scenarios.
One email per alert, topics in terms nobody recognised, and an archive most people never found. Funding officers felt overwhelmed.
9 interviews on both sides, a five-phase journey map, and a redesign of the three phases that hurt most.
A tested prototype: preferences by alert type, a weekly digest, deadline-first urgent alerts and a deadline-led portal.
Research Alerts pushes funding opportunities, awards, workshops and events to subscribers by email. Every published alert is also kept on a website, where signed-in users can search the collection. The system is a custom build that predates the team that supports it.
Six weeks, 9 interviews: 4 people who create and approve alerts, 5 who receive them. An affinity board on one side, an empathy map on the other.
“There is no way! I have no time or no need to go back to the portal.”End user
“I don’t even know you can search on RA.”End user
“I know I can filter, but I just signed up for everything.”End user
“I bet 99% or 95% of users select all of these (tags).”Approver
“Everything captured in same platform… it gets to be a little bit overwhelming.”End user
“I rely on email. But I get too many emails.”Research approver
Difficulty to highlight important information and timelines.Creators, interview note
“It’s sometimes hard to skim for the important details.”End user
Search by author returns the most recent entry only.Research approver, interview note
Search not intuitive, never searched before, not aware there is a search function.End user, interview note
Emails are great and short, not too much but enough information.End user, interview note
A Business Model Canvas and a Value Proposition Canvas helped us see Research Alerts from VPRI’s side and decide who to design for.
How other research-alert and funding-opportunity sites handle filtering and layout.
Key decision
The brief
Fix the website
Improve the website where people search the collection of published alerts.
What research showed
People live in email
Most end users never open the website. They read subject lines, skim, and forward what applies. The pain sits in sign-up, email volume and relevance.
What we decided
Redesign three stages, email first
Focus on Subscription, Notification and Usage, which hold six of the seven pain points we mapped. The email becomes the entry point, and the portal is where an email leads.
Motivation (finding the service) and Exit (unsubscribing) stayed out of scope. In the As-Is map, Exit was the one phase with no pain points.
We built the persona around the main user group, end users. Rita Amato is a research funding officer: she checks email every day, forwards the alerts that apply to faculty, and has no time to log in to another system.
“I wish there was a way to separate grants, awards and funding opportunities from other alerts.”
Rita Amato, Research Funding Officer, 45, Toronto. Persona.
How Rita feelsOverwhelmedConfusedUncertain
We mapped what Rita does, feels and thinks across five phases, along with pain points and opportunities for VPRI. Between the first and second playback we rebuilt the map: the steps became one continuous path that shows Rita’s daily email loop and where usage branches off, touchpoints are named by service instead of device, and VPRI’s goals and opportunities were condensed into one row of opportunities.
| Phase | How Rita feels | Pain points | Opportunities for VPRI |
|---|---|---|---|
| Motivation | Curious, excited | Difficulty finding the Research Alerts page | Make the service easier to find |
| Subscription In scope | Hopeful, confused | Little diversity of tags for setting preferences | Let users customise their alert preferences |
| Notification In scope | Overwhelmed, intrigued, satisfied | Hard to skim and judge the relevance of alerts. The influx of alerts can be overwhelming. | Send only relevant alerts. Make the key content clear in the email. Make users aware they can manage preferences. |
| Usage In scope | Overburdened, informed, uncertain | Unaware the portal exists. Search and filter aren’t intuitive. No preview of the alert summary. | Help users reach relevant alerts when they need them. Raise awareness of what the portal can do. |
| Exit | Relaxed | None found | Let users say why they unsubscribe |
We ideated broadly, including a few deliberately absurd ideas, then plotted every idea by its value for users against its feasibility for our team.
What made the cut. High-value ideas became the core of the redesign: alert tags you can add and remove, a full list of programs and faculties, reminders for urgent alerts, a calendar of deadlines, a weekly digest, condensed alerts with a “see more” link, and one-click navigation.
What didn’t. Ideas in the low-value, low-feasibility corner, such as an Outlook pop-up and saving alerts as favourites, stayed out of the prototype.
We sketched four concepts and walked the VPRI team through them. Their feedback went into the high-fidelity designs.
We ideated broadly, plotted every idea by value for users against feasibility for our team, and walked VPRI through four sketched concepts before going hi-fi. Tags, a weekly digest, urgent reminders and a calendar of deadlines made the cut; an Outlook pop-up and favourites didn’t.
Each change answers a specific finding. Flip between before and after.
02Subscription
Sketch: onboarding and preferences, then prototype, then tested in task 1
Before: checkbox lists for discipline, function and role.After: alert-type tabs, frequency, program and faculty.
03Notification
Sketch: redesigned emails, then prototype, then tested in task 2
Before: one email per alert, deadlines buried in the text.After: tags, a deadlines box and one-click navigation.
04Usage
Sketches: dashboard and calendar, then prototype, then tested in tasks 1 and 2
Before: a table whose filters only covered the current page.After: upcoming deadlines, keyword search and tagged alerts.
We wrote task scenarios around Rita’s day and storyboarded every screen and click participants were expected to make.
Task 1
The user learns about Research Alerts and goes to its login and sign-up page.
Task 2
Having just signed up, the user now receives the daily digest by email.
Prototyping and usability testing fed the To-Be experience map. Across subscription, notification and usage, the low points of the As-Is journey gave way to positive moments.
Emotional low points between subscription and usage, from 4 in the As-Is map to 0 in the To-Be map.
Confused, overwhelmed, overburdened, uncertainHopeful, easy, relaxed, satisfied
We closed the project in a final discussion with VPRI covering the move from As-Is to To-Be and the future scope for redesigning Research Alerts.
The brief pointed at the portal. Interviews showed people live in their inbox, so we designed there first and treated the portal as the place an email leads.
Plotting ideas by user value and team feasibility kept the redesign realistic for a custom, 9-year-old system.
Each change in the prototype traces back to a pain point or a participant quote, and the main flows were checked in usability tasks.
Research Alerts Client project for the Division of the Vice-President, Research & Innovation (VPRI), University of Toronto. INF2192 Representing UX, Winter 2022. Team Garima Batra, Sahil Faraz, Zimeng Li, Navid Rahman, Mona Abou Swaid, Nan Wang.
Names and email addresses of VPRI staff, alert authors and participants are blurred in all screenshots.
Mixed reality product design, University of Toronto, 2021
A quiet office you can wear. An app concept for mixed reality glasses that helps students and desk workers focus in busy, shared places.
From 6 research sources to 11 hi-fi screens, with a background chosen by 14 people and 3 ways to interact.
Students and desk workers lose focus to the people around them and the phone on the table. Home, libraries and cafés all came with crowds.
A team of three: six sources and a storyboard. On my own: UI sketches, a user interview, a 14-person preference test and the hi-fi UI.
11 hi-fi screens for MR glasses: two modes, custom partitions, whitelisting, a people indicator and a grayscale fade for breaks.
Team, INF2310
Solo, INF2191
A year before this project, in an individual exploration, I interviewed peers and family and collected responses through an Instagram story, about two dozen in all. Distraction while working stood out as a real and recurring problem.
“I have the superpower to get distracted anywhere!”
Michael is 28, a student in Toronto with high tech literacy. Assignments have piled up, and the people around him and the phone next to him keep pulling him away.
For Aug-Fice we kept the same audience, students and people with desk jobs, and focused on visual distractions: people moving around them and the phone on the table.
I wrote up early ideas for tackling distraction with mixed reality in a Medium article.
Team
We read studies on virtual partitions and VR workspaces, the Pomodoro technique, diminished reality and digital wellbeing. Each one turned into something the user can control.
Lee et al., 2019Partitioning open-plan workspaces via AR. 48 participants, 11 expert interviews.
Any partition beat none. People preferred semi-transparent partitions, and experts said transparency and coverage should come first.
Adjustable transparency and distance
Lee & Bianchi, 2020YourSphere. 18 interaction designers.
People imagined partitions as virtual panels, shapes extruded from drawn lines, extended screens, or spheres around them.
Shapes beyond a rectangle: cubicle, sphere, blinds
Ruvimova et al., 2020“Transport Me Away,” CHI. 35 participants.
Flow was highest in a closed office. People preferred it and a VR beach, and strongly disliked open offices.
Calm 360° environments to work in
Cirillo, 2006The Pomodoro Technique.
25-minute work periods with 5-minute breaks, a planning step first, and breaks that truly detach you from the task.
Focus mode with sets, a to-do list and break transitions
Mori et al., 2017A survey of diminished reality.
Real objects can be diminished, seen through, replaced or inpainted.
Hide the phone screen, ads and distracting websites
GoogleDigital Wellbeing experiments.
Tools that make phone use visible or cut apps and notifications down to the essentials.
Grayscale screens to weaken the pull of apps
Concept, team
Focus runs Pomodoro sets with a to-do list. Zen is untimed. In both, a virtual partition hides the room, and the glasses hide the phone screen and distracting sites.
TransparencyMedium
ShapeBlinds
People indicatorOn
Public: medium transparency with blinds, so you’re hidden but not cut off.
1Distracted in a shared space: people talking, a cat, a phone.
2Puts on the MR headset.
3Picks a preset (Office, Home or Public space) with gaze and air tap.
4Sets the cubicle’s shape, background and distance.
5Chooses Pomodoro over Zen.
6Sets work and break times and imports a to-do list.
7The cubicle fades in and the room fades out. The phone rings.
8The phone shows the call isn’t important. At the break, the partition clears and screens turn gray.
Turning point
“It’s Steps 7 and 8 where your real ideas are.”
Prof. Tony Tang, on our storyboard
He called the setup steps boilerplate and asked what happens at the boundary between the user and the room: “If I step close to you, am I still diminished? At what point do I become something that you cannot ignore?” The hi-fi design builds out that boundary.
New
Nearby people appear as markers in three distance bands whenever the partition is more than half opaque. Brighter means closer.
Built out
Step 8’s “not important” message became a screen for choosing the contacts and apps that can still reach you.
Built out
Step 8’s gray screens became an opt-in grayscale fade, with the partition clearing as a set ends.
New
Each mode remembers its last settings, and feature intros only appear on first run.
Still open: how people nearby can tell whether it’s OK to interrupt. The interview suggested a plain desk sign.
Solo
I sketched every screen on paper and revised the sketches in red pen before moving to Figma.
Three kinds of feedback: a storyboard critique, a flow interview and a background preference test.
Peer review, 2 classmates
Both liked pairing productivity with wellbeing. They asked how audio distractions would be handled, for a privacy explanation before first use, for one storyline per mode, and for more annotations.
Critique, Prof. Tony Tang
Praised the grounding in research and called steps 7 and 8 the most promising. He asked how people sitting nearby should be treated, and how they can tell whether to interrupt.
User interview, 1 participant Solo
28, a tech enthusiast with about ten VR sessions and no hands-on AR. We didn’t cover gestures because the participant didn’t know the common AR ones.
Background preference test Solo
14 people saw the mode screen on different backgrounds and voted for the ones they liked. Each person could pick more than one.
C was picked by 8 of the 14 people and became the background of the final UI.
| What we heard | From | How the design responds |
|---|---|---|
| “An all-white UI could feel monotonous and claustrophobic.” | Interview | A dark, textured UI with bright colours used sparingly, which also eases VR discomfort. Background C won the vote. |
| “Hiding the phone fails when it rings.” | Interview | Whitelisting stays in the app: choose the contacts and apps that can reach you. Showing those alerts inside the glasses is left to the headset’s OS. |
| “Privacy features feel scary.” “Explain privacy before first use.” | Interview, peer review | The people indicator and grayscale fade are off by default, and each gets a one-screen explanation with a No / Yes choice. |
| “If I step close to you, am I still diminished?” | Professor | The people indicator shows nearby people in three distance bands when the partition is more than half opaque. |
| “Add more annotations.” | Peer review | Every screen on the hi-fi storyboard carries its own annotation. |
Still open: audio distractions, a storyline for Zen mode, letting others know you’re focusing, and a shared “virtual cabin” for teams.
PlannedNext round
5 to 8 students and desk workers, mixing people new to AR with people who have used it.
First a click-through of the Figma prototype on a laptop, to test the flow and wording. Then a headset session, to test gestures, comfort and the partition.
Task success, time on task and errors; a one-question ease rating per task; SUS at the end; simulator sickness and neck comfort in the headset.
| Task | Screens it covers |
|---|---|
| Start a Focus session of four 25-minute sets with 5-minute breaks. | Choose a mode, Customize session |
| Set up the Public preset with a rectangle partition 5 ft away and the people indicator on. | Choose your environment, Customize Public, People indicator |
| Let calls from one contact through while you work. | Whitelisting |
| Import your to-do list, start working and tick off a task. | Prepare to-do list, Work |
| Notice the break, then get back to work. | Break, Work |
Visual design, solo
I collected references in a Miro moodboard, then wrote a style guide in Figma covering colour, contrast, type, the logo, motion, and layout rules for head tracking.
Aa
Charcoal#393D36, background
Aa
Green#8BC34A, primary
5.27:1 on charcoal
Aa
Gold#D9C37E, breaks
6.36:1 on charcoal
Aa
White at 75%#FFFFFF, primary text
11.08:1 on charcoal
The logo is two overlapping circles: the real and virtual worlds, with mixed reality where they meet. They also stand for the two modes.
Bright backgrounds add to VR sickness, so whites and bright colours are used sparingly.
Sessions and breaks fade in and out with saturation and dissolve transitions, so eyes can adjust.
Targets are big, and the most-used ones sit in the centre to limit neck movement.
Colour pairs were checked with the Stark plugin; green and gold on charcoal both pass WCAG AA.
Avenir Next LT Pro for headings, Source Sans Pro for body text and small sizes.
Corner brackets let users resize any window by hand gesture, and the lock pins the UI in place during a quick interruption.
Final design, solo
Eleven hi-fi screens in Figma, built on the team’s user flow. Step through them, or use the arrow keys.
1 of 11
Design decision
Assumes to-do lists sync from a web app, keyboards pair with the glasses (or a virtual keyboard like Immersed), cloud services connect to the app, and back and home gestures are standard.
Transparency, distance and shape came straight from the partition studies, and 360° environments from the VR workspace study.
The moments that matter are when the real world gets back in: someone walks up, an important call comes in, a set ends.
Anything that watches the room starts switched off and explains itself first. The interview showed how quickly trust can drop.
Aug-Fice, INF2310 (guided by Prof. Tony Tang) and INF2191, University of Toronto, Fall 2021. Concept team: Sahil Faraz, Grace Chen, Sabrina Yeasmin.