Quiet coaching. Real progress.
Designing AI-integrated gym equipment where feedback stays quiet, control stays with the user, and everything stays private from the room.
Designing AI-integrated gym equipment where feedback stays quiet, control stays with the user, and everything stays private from the room.
78%
Want private coaching
Want private coaching
82%
Want AI override
Want AI override
75%
Prefer haptic over screen feedback
75%
Prefer haptic over screen feedback
OVERVIEW
OVERVIEW
AURA is an AI-integrated gym equipment system that delivers real-time, body-aware coaching without putting anyone on display. Feedback lives in the hardware, ambient LED bands, localized haptics, adaptive resistance, not on a public screen.
The intelligence stays in the background. The user stays in charge.
AURA is an AI-integrated gym equipment system that delivers real-time, body-aware coaching without putting anyone on display. Feedback lives in the hardware, ambient LED bands, localized haptics, adaptive resistance, not on a public screen.
The intelligence stays in the background. The user stays in charge.
My Role
My Role
Product Designer
Product Designer
UX Research and UI Design
UX Research and UI Design
Timeline
Timeline
16 weeks
16 weeks
Team
Team
4 members, Arizona State University
4 members, Arizona State University
Highlight
Building an AI gym system where the equipment does the coaching and the room sees nothing.
Highlight
Building an AI gym system where the equipment does the coaching and the room sees nothing.
PROBLEM
PROBLEM
Smart gyms got loud. Real users got quieter.
Connected fitness scaled fast, but the design drifted away from how people actually feel in a shared gym.
Connected fitness scaled fast, but the design drifted away from how people actually feel in a shared gym.
01
The equipment treats the user as a performer.
Big screens. Public dashboards. Voice prompts anyone nearby can hear. The interfaces assume you're either home alone or proud to be watched. Most users we spoke to were neither. Beginners felt exposed and rushed. Even experienced lifters said the constant performance framing pulled them out of their own body.
01
The equipment treats the user as a performer.
Big screens. Public dashboards. Voice prompts anyone nearby can hear. The interfaces assume you're either home alone or proud to be watched. Most users we spoke to were neither. Beginners felt exposed and rushed. Even experienced lifters said the constant performance framing pulled them out of their own body.
02
The AI gives advice, not control.
Connected systems are getting smarter, but smarter doesn't mean more trusted. Every persona we built pushed back on automation they couldn't turn off. A system that adjusts your resistance or flashes a warning without a clear override feels presumptuous.
02
The AI gives advice, not control.
Connected systems are getting smarter, but smarter doesn't mean more trusted. Every persona we built pushed back on automation they couldn't turn off. A system that adjusts your resistance or flashes a warning without a clear override feels presumptuous.
THE CHALLENGE
THE CHALLENGE
"How might we provide adaptive, autonomy-respecting coaching in shared gym environments, without making the user feel watched?"
"How might we provide adaptive, autonomy-respecting coaching in shared gym environments, without making the user feel watched?"
A gap existed between home fitness tech and commercial gym consoles.
One assumes you're alone. The other treats every user the same. Neither was built for the emotional reality of a shared gym.
One assumes you're alone. The other treats every user the same. Neither was built for the emotional reality of a shared gym.
HOME-CONNECTED HARDWARE
HOME-CONNECTED HARDWARE
Peloton · Tonal · Mirror icons Great coaching, big screen, assumes a private room.
COMMERCIAL GYM CONSOLES
COMMERCIAL GYM CONSOLES
Technogym · Life Fitness icons Tracks data. Treats every user identically.
AURA
AURA
Personality-first. Privacy-first icon Ambient signals on the equipment. Intelligence in the background. Control in the user's hands.
HOME-CONNECTED HARDWARE
Peloton · Tonal · Mirror icons Great coaching, big screen, assumes a private room.
COMMERCIAL GYM CONSOLES
Technogym · Life Fitness icons Tracks data. Treats every user identically.
AURA
Personality-first. Privacy-first icon Ambient signals on the equipment. Intelligence in the background. Control in the user's hands.
No solution balances coaching, privacy, and user control.
No solution balances coaching, privacy, and user control.
RESEARCH
RESEARCH
Who we talked to
Surveys for patterns, interviews for emotion, observation for checking what people said against what they did.
Surveys for patterns, interviews for emotion, observation for checking what people said against what they did.
50+
Survey respondents
50+
Survey respondents
3
External in-depth interviews
3
External in-depth interviews
1
In-class mock interview
1
In-class mock interview
What we heard, again and again
4 signals shaped every design call after this.
4 signals shaped every design call after this.
78%
Rated private coaching as very or extremely important.
78%
Rated private coaching as very or extremely important.
82%
Wanted the ability to override AI suggestions.
82%
Wanted the ability to override AI suggestions.
71%
Expressed concern about public correction or visible stats.
71%
Expressed concern about public correction or visible stats.
75%
Preferred haptic feedback over visible screen prompts.
75%
Preferred haptic feedback over visible screen prompts.
01 - What we learned
Internal doubt beats external judgment.
Internal doubt beats external judgment.
People fear doing something wrong more than being seen doing it.
People fear doing something wrong more than being seen doing it.
Confidence isn't stable.
Confidence isn't stable.
The same person can feel at ease on one machine and frozen on the next.
The same person can feel at ease on one machine and frozen on the next.
Trust in automation comes down to three things
Trust in automation comes down to three things
Does it feel safe? does it match my mood? and can I turn it off?
Does it feel safe? does it match my mood? and can I turn it off?
02 - Competitive teardown
drag me
The market splits into three categories, and none of them were built for the emotional reality of a shared gym. Home hardware assumes a private room. Gym consoles treat every user identically. Wearables track outcomes but can't correct your form mid-set.
The market splits into three categories, and none of them were built for the emotional reality of a shared gym. Home hardware assumes a private room. Gym consoles treat every user identically. Wearables track outcomes but can't correct your form mid-set.
03 - Three personas
Maya Chen
24, Beginner, Self-conscious
I want to know I'm doing it right without everyone knowing I'm new
Behavioral tags:
• Avoids asking for help
• Copies others instead of asking
• Looks calm, feels uncertain
Maya Chen
24, Beginner, Self-conscious
I want to know I'm doing it right without everyone knowing I'm new
Behavioral tags:
• Avoids asking for help
• Copies others instead of asking
• Looks calm, feels uncertain
Daniel Wright
Experienced lifter, Mood-driven
Don't coach me. Just have my back when it matters
Behavioral tags:
• Adjusts weights based on how he feels
• Ignores coaching prompts
• Pushes through fatigue manually
Daniel Wright
Experienced lifter, Mood-driven
Don't coach me. Just have my back when it matters
Behavioral tags:
• Adjusts weights based on how he feels
• Ignores coaching prompts
• Pushes through fatigue manually
Meera Austin
Semi-confident, Learns by trying
I just need to know when I'm close
Behavioral tags:
• Pauses to evaluate feedback
• Gains confidence when feedback is clear
• Loses trust when instructions feel vague
Meera Austin
Semi-confident, Learns by trying
I just need to know when I'm close
Behavioral tags:
• Pauses to evaluate feedback
• Gains confidence when feedback is clear
• Loses trust when instructions feel vague
Different goals, different fitness levels, same underlying ask: guidance that doesn't expose them.
Different goals, different fitness levels, same underlying ask: guidance that doesn't expose them.
04 - From data to design priorities
04 - From data to design priorities
HMW provide form guidance without asking for help
HMW provide form guidance without asking for help
HMW provide form guidance without asking for help
HMW adapt equipment to mood and energy
HMW adapt equipment to mood and energy
HMW adapt equipment to mood and energy
HMW create feedback that builds confidence progressively
HMW create feedback that builds confidence progressively
HMW create feedback that builds confidence progressively
HMW track progress without overwhelming the user
HMW track progress without overwhelming the user
HMW track progress without overwhelming the user
HMW preserve autonomy while offering intelligent coaching
HMW preserve autonomy while offering intelligent coaching
HMW preserve autonomy while offering intelligent coaching
Privacy-first
Privacy-first
Override-first
Override-first
Calm-tech
Calm-tech
The reframe that stuck: moving from "how do we improve gym equipment" to "how do we support confidence in shared spaces without making people feel watched."
The reframe that stuck: moving from "how do we improve gym equipment" to "how do we support confidence in shared spaces without making people feel watched."
05 - How the system responds
Research gave us three rules: feedback only when needed, the user always in control, and nothing visible to the room. The user flow is where those rules became a decision architecture.
Research gave us three rules: feedback only when needed, the user always in control, and nothing visible to the room. The user flow is where those rules became a decision architecture.
drag me
The branching node at "Posture Deviation Detected?" is the most important decision in the entire system. If no deviation is detected, AURA stays silent. If one crosses the threshold, a private cue fires, and the user decides what happens next. That single fork shaped every screen that came after it.
The branching node at "Posture Deviation Detected?" is the most important decision in the entire system. If no deviation is detected, AURA stays silent. If one crosses the threshold, a private cue fires, and the user decides what happens next. That single fork shaped every screen that came after it.
06 - Built with Figma MCP, then refined by hand.
The design system was generated using Figma's MCP server: tokens, components, and layout rules created programmatically, then reviewed and iterated manually where the output fell short.
Why MCP first, then manual
Why MCP first, then manual
MCP generated a consistent foundation fast: tokens, type scale, button variants, color system. But the output needed human judgment: the toggle switch colors clashed with the brand palette (caught in heuristic eval), the body-map overlay needed custom layering MCP couldn't infer, and the semantic color for deviation severity was reusing the primary accent, blurring the line between interactive elements and status indicators. We fixed all three manually.
MCP generated a consistent foundation fast: tokens, type scale, button variants, color system. But the output needed human judgment: the toggle switch colors clashed with the brand palette (caught in heuristic eval), the body-map overlay needed custom layering MCP couldn't infer, and the semantic color for deviation severity was reusing the primary accent, blurring the line between interactive elements and status indicators. We fixed all three manually.
The takeaway: MCP is a scaffolding tool, not a finishing tool. It got us to 80% in a fraction of the time. The last 20% - the part that makes a system feel intentional - still required a designer.
The takeaway: MCP is a scaffolding tool, not a finishing tool. It got us to 80% in a fraction of the time. The last 20% - the part that makes a system feel intentional - still required a designer.
PRODUCT
PRODUCT
The final app is built around three ideas.
The final app is built around three ideas.
Each solves a different piece of the same problem. Together, they behave like one tool, not three features stitched into an app.
Each solves a different piece of the same problem. Together, they behave like one tool, not three features stitched into an app.
1.
1.
The equipment is the interface, not the screen.
The equipment is the interface, not the screen.
Feedback lives in the hardware: LED bands, haptics, adaptive resistance. The app stays private. Nothing broadcasts to the room. A beginner feels a wrist cue. An experienced lifter feels a vibration when form drifts.
Feedback lives in the hardware: LED bands, haptics, adaptive resistance. The app stays private. Nothing broadcasts to the room. A beginner feels a wrist cue. An experienced lifter feels a vibration when form drifts.
2.
2.
The body map is the dashboard.
The body map is the dashboard.
A single body-map figure shows muscle engagement, resistance zones, risk areas. That's the whole dashboard, and it stays personal. Leaderboards became trend lines. Public scores became private streaks.
A single body-map figure shows muscle engagement, resistance zones, risk areas. That's the whole dashboard, and it stays personal. Leaderboards became trend lines. Public scores became private streaks.
3.
3.
Override is a first-class feature, not a setting.
Override is a first-class feature, not a setting.
Every adaptive change is cancelable. Every scan is skippable. Every alert is dismissible. The override isn't buried in preferences, it's a visible button on every screen that uses it.
Every adaptive change is cancelable. Every scan is skippable. Every alert is dismissible. The override isn't buried in preferences, it's a visible button on every screen that uses it.
Visuals of AURA
TESTING
TESTING
What the test actually told us.
Low-fidelity evaluation across both critical flows. Figma prototype for the Scan System, paper for the Motivation System. Mixing formats was intentional: screen-state testing vs. emotional-flow iteration.
Low-fidelity evaluation across both critical flows. Figma prototype for the Scan System, paper for the Motivation System. Mixing formats was intentional: screen-state testing vs. emotional-flow iteration.
OVERVIEW STATS:
OVERVIEW STATS:
3 participants (1 mock, 2 external)
3/4 screens worked first time
8 tasks total (4 per feature)
1 screen rebuilt from scratch
3 participants (1 mock, 2 external)
8 tasks total (4 per feature)
1 screen rebuilt from scratch
3/4 screens worked first time
3 participants (1 mock, 2 external)
1 screen rebuilt from scratch
8 tasks total (4 per feature)
3/4 screens worked first time
TASKS
TASKS
RESULT
RESULT
Setup and calibration
Setup and calibration
Read as natural
Read as natural
Exercise selection and live scan
Exercise selection and live scan
Body-map read instantly
Body-map read instantly
Reading live feedback
Reading live feedback
Threshold alerts landed
Threshold alerts landed
End-of-session review
End-of-session review
Streaks read as effort, not score
Streaks read as effort, not score
Starting a session
Starting a session
Worked after Start Now button revision
Worked after Start Now button revision
Your Circle (social activity)
Your Circle (social activity)
Understood immediately
Understood immediately
Consistency feedback
Consistency feedback
Failed. Rebuilt with metric tiles
Failed. Rebuilt with metric tiles
Profile view
Profile view
Streak displays read without labels
Streak displays read without labels
What users told us
Two confirmed the design. One forced a rebuild.
Two confirmed the design. One forced a rebuild.
The Concept Landed
Tester 1
24 Feb 2026
"It reminded me of setting up Face ID. I didn't think about it, I just did it."
Confirmed the calm-tech direction.
The Concept Landed
Tester 1
24 Feb 2026
"It reminded me of setting up Face ID. I didn't think about it, I just did it."
Confirmed the calm-tech direction.
The Concept Landed
Tester 1
24 Feb 2026
"It reminded me of setting up Face ID. I didn't think about it, I just did it."
Confirmed the calm-tech direction.
Your Circle landed without instruction.
Tester 2
18 Mar 2026
"I got it without anyone telling me what it was."
Validated the "consistency over competition" call.
Your Circle landed without instruction.
Tester 2
18 Mar 2026
"I got it without anyone telling me what it was."
Validated the "consistency over competition" call.
Your Circle landed without instruction.
Tester 2
18 Mar 2026
"I got it without anyone telling me what it was."
Validated the "consistency over competition" call.
The consistency screen failed the user it was built for.
Tester 3 (Maya-fit)
9 Apr 2026
"I don't know what this is asking me to do."
Rebuilt the same week - text-only replaced with three metric tiles.
The consistency screen failed the user it was built for.
Tester 3 (Maya-fit)
9 Apr 2026
"I don't know what this is asking me to do."
Rebuilt the same week - text-only replaced with three metric tiles.
The consistency screen failed the user it was built for.
Tester 3 (Maya-fit)
9 Apr 2026
"I don't know what this is asking me to do."
Rebuilt the same week - text-only replaced with three metric tiles.
Before / After / Final
Two confirmed the design. One forced a rebuild.
Two confirmed the design. One forced a rebuild.
Iteration 1 - Before
text-only consistency screen. No metric tiles, no structured feedback. The user couldn't interpret what the screen was communicating or what action came next.
Iteration 2 - After
three metric tiles (Duration, Calories, Streaks), rewritten motivation overlay referencing actual workout count, visually distinct action buttons (Make it Happen / Maybe Later).
Iteration 3 - Final
Same three-tile structure, dark theme, acid-lime accents, personalized copy.
Iteration 1 - Before
text-only consistency screen. No metric tiles, no structured feedback. The user couldn't interpret what the screen was communicating or what action came next.
Iteration 2 - After
three metric tiles (Duration, Calories, Streaks), rewritten motivation overlay referencing actual workout count, visually distinct action buttons (Make it Happen / Maybe Later).
Iteration 3 - Final
Same three-tile structure, dark theme, acid-lime accents, personalized copy.
IMPACT
IMPACT
The project moved the design in real ways, even without shipping.
Outcomes grouped by what they actually proved
Outcomes grouped by what they actually proved
01. CLARITY -
01. CLARITY -
Did the design land with real users?
Did the design land with real users?
3/4 core screens
read on first contact: Setup, Body Map, Your Circle, Streaks.
3/4 core screens
read on first contact: Setup, Body Map, Your Circle, Streaks.
1 rebuilt from scratch
after failing the user it was built for.
1 rebuilt from scratch
after failing the user it was built for.
02. VALIDATION -
02. VALIDATION -
Did the concept hold up?
Did the concept hold up?
Research-backed design system with
every rule rooted in user insights.Override on every adaptive change. Privacy on every screen. Haptics over voice. Alerts only when they matter.
Research-backed design system with
every rule rooted in user insights.Override on every adaptive change. Privacy on every screen. Haptics over voice. Alerts only when they matter.
8 flows shipped at Hi-Fi
on a unified design system.
8 flows shipped at Hi-Fi
on a unified design system.
Critiques caught the design
drifting toward gamification twice. Both would have contradicted the premise.
Critiques caught the design
drifting toward gamification twice. Both would have contradicted the premise.
03. STRATEGY -
03. STRATEGY -
Did the bet pay off?
Did the bet pay off?
Validated a category.
A privacy-first, autonomy-respecting AI coaching system for shared gyms. No existing product occupies this space.
Validated a category.
A privacy-first, autonomy-respecting AI coaching system for shared gyms. No existing product occupies this space.
Scoped a roadmap.
Physical prototype, longitudinal pilot, and equipment partnership. Surfaced a design principle. From exposure to confidence, through privacy, autonomy and support.
Scoped a roadmap.
Physical prototype, longitudinal pilot, and equipment partnership. Surfaced a design principle. From exposure to confidence, through privacy, autonomy and support.
LEARNING
LEARNING
What This Project Taught Me
What This Project Taught Me
Lessons learned about user behavior, product strategy, and designing within constraints.
Emotional safety is a usability requirement.
The consistency screen failed on comprehension. What it cost was confidence. In any AI tool where people feel exposed, clarity at vulnerable moments is what turns uncertainty into trust.
Emotional safety is a usability requirement.
The consistency screen failed on comprehension. What it cost was confidence. In any AI tool where people feel exposed, clarity at vulnerable moments is what turns uncertainty into trust.
Autonomy is what makes AI features trustworthy.
People accept a lot of automation as long as they can opt out without digging. The override isn't a fallback - it's the thing that earns trust for everything else.
Autonomy is what makes AI features trustworthy.
People accept a lot of automation as long as they can opt out without digging. The override isn't a fallback - it's the thing that earns trust for everything else.
Stillness beats noise.
The biggest wins came from saying less. Voice prompts went out. Big screens went out. Constant feedback went out. The system that worked intervened only when it had to.
Stillness beats noise.
The biggest wins came from saying less. Voice prompts went out. Big screens went out. Constant feedback went out. The system that worked intervened only when it had to.
Internal doubt beats external judgment.
We assumed users feared what others thought. Research said the opposite - they feared what they thought of themselves. That shift rewrote half our copy.
Internal doubt beats external judgment.
We assumed users feared what others thought. Research said the opposite - they feared what they thought of themselves. That shift rewrote half our copy.
What comes next
Each phase of AURA was scoped to validate the next one.
Each phase of AURA was scoped to validate the next one.
Next Steps - short term:
a. Build a physical prototype, squat-rack integration as the first build.
b. Expand user pool: older adults, injury recovery, documented gym anxiety.
c. Run a 4-6 week longitudinal pilot to test whether consistency features change behavior or novelty is doing the work.
Next Steps - short term:
a. Build a physical prototype, squat-rack integration as the first build.
b. Expand user pool: older adults, injury recovery, documented gym anxiety.
c. Run a 4-6 week longitudinal pilot to test whether consistency features change behavior or novelty is doing the work.
Future Ideas - long term:
a. AURA as an ambient gym layer, multiple equipment pieces recognizing the user across the floor without identifying them to the room.
b. Equipment partnerships with Technogym or Life Fitness to retrofit existing consoles.
c. Calm-tech feedback patterns applied beyond fitness - healthcare, education, financial coaching.
Future Ideas - long term:
a. AURA as an ambient gym layer, multiple equipment pieces recognizing the user across the floor without identifying them to the room.
b. Equipment partnerships with Technogym or Life Fitness to retrofit existing consoles.
c. Calm-tech feedback patterns applied beyond fitness - healthcare, education, financial coaching.
"AURA was built around one hypothesis: that an AI system can be smart without being loud, supportive without being intrusive, and personal without being performative. So far, the evidence is promising."
"AURA was built around one hypothesis: that an AI system can be smart without being loud, supportive without being intrusive, and personal without being performative. So far, the evidence is promising."
NEXT PROJECT
NEXT PROJECT
©2026 Rishi Kataria • Arizona
©2026 Rishi Kataria • Arizona
©2026 Rishi Kataria • Arizona