Hit
T
your keyboard for express version →
•
2025
Scaling AI-Generated Imagery: An 83% Faster Production System, with 100% imagery coverage.
•
2025
Scaling AI-Generated Imagery: An 83% Faster Production System, with 100% imagery coverage.
MY ROLE
Lead UX Design
TIMELINE
6 Months
RESPONSIBILITIES
UX Research
UX Research, Stakeholder Presentations, Prototyping, User Testing, UI Design.
Stakeholder Presentations
Prototyping
User Testing
UI Design
A talent acquisition platform seeking a compelling UX approach and immersive experience.
BMW STEP is amongst the most extensive technician training schemes across North America. They are recognized for their practical methodology and thorough enhancement of technical skills. Yet, their official website seemed antiquated and overloaded with content, lacking a narrative to support its value proposition. As a result, the site faced increased bounce rates, diminished traffic, and reduced page duration.
TL;DR
Rebuilt an image tool capped at 80% inventory coverage into a system that generates unlimited GenAI asset combinations, reaching 100% coverage and cutting setup time from 5 minutes to 1:15.
TL;DR
Rebuilt an image tool capped at 80% inventory coverage into a system that generates unlimited GenAI asset combinations, reaching 100% coverage and cutting setup time from 5 minutes to 1:15.
PROBLEM
A car without a photo is a car losing money
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
SOLUTION
Scale coverage without breaking what people already knew
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
PROBLEM
A car without a photo is a car losing money
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
SOLUTION
Scale coverage without breaking what people already knew
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
MY ROLE
Senior Product Designer
TIMELINE
3 Months
TEAM
PM, 2 Engineers and me
RESPONSIBILITIES
UX Design
Stakeholder Presentations
Prototyping
User Testing
TL;DR
Rebuilt a GenAI tool capped at 80% inventory coverage into a system that generates unlimited asset combinations, reaching 100% coverage and cutting setup time from 5 minutes to nearly 1 minute.
TL;DR
Rebuilt an image tool capped at 80% inventory coverage into a system that generates unlimited GenAI asset combinations, reaching 100% coverage and cutting setup time from 5 minutes to nearly 1 minute.
PROBLEM
A car without a photo is a drain on money
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
SOLUTION
Scale coverage without breaking what people already knew
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
PROBLEM
A car without a photo is a drain on money
Every day a vehicle sits unlisted counts against it: financing, insurance and depreciation pile up, the metric dealers track as days on lot. The previous tool covered only 80% of inventory, since it generated a single image configuration per vehicle. What it did cover was painful to produce, with every step packed into a cramped right panel and no full view of the result.
SOLUTION
Scale coverage without breaking what people already knew
A config is a saved AI image template: background, angle and lighting, that can be applied and integrated to any VIN (vehicle). The mandate was to go from one config for the entire inventory to unlimited config-to-VIN combinations, without forcing a new mental model onto existing users. Speed had to scale with coverage: the old flow took about 5 minutes per set, the new one takes one minute.
IMPACT
IMPACT
IMPACT
83%
faster task completion
83%
faster task completion
83%
faster task completion
200+ US dealers
using this solution for speed to market
200+ US dealers
using this solution for speed to market
200+ US dealers
using this solution for speed to market
100%
inventory coverage with new tool
100%
inventory coverage with new tool
100%
inventory coverage with new tool
Main Goals
Main Goals
Enable scalable configuration logic across growing VIN complexity
Reduce manual mapping errors
Increase governance without sacrificing flexibility
Future-proof the system for market-level variations
Research
Research
For this project, I relied on competitive and comparative analysis in order to gather UX standards that support the project goals, and to dovetail them in a seamless and elegant way.
For this project, I relied on competitive and comparative analysis in order to gather UX standards that support the project goals, and to dovetail them in a seamless and elegant way.
For this project, I relied on competitive and comparative analysis in order to gather UX standards that support the project goals, and to dovetail them in a seamless and elegant way.
UX Strategy
UX Strategy
Instead of extending the single-config model, we re-architected the system to support:
Multiple configurations per VIN
Scoped rule-based application
Permission-based overrides
Visualized impact before deployment
This shifted the platform from static assignment to dynamic logic orchestration. The design challenge became one of containment — shaping complexity instead of hiding it.
Instead of extending the single-config model, we re-architected the system to support:
Multiple configurations per VIN
Scoped rule-based application
Permission-based overrides
Visualized impact before deployment
This shifted the platform from static assignment to dynamic logic orchestration. The design challenge became one of containment — shaping complexity instead of hiding it.
DESIGN EVOLUTION
DESIGN EVOLUTION
EVOLUTION
1
PoC Experimental Version
First version had all controls in one side panel, making the config experience painful and convoluted
1
PoC Experimental Version
First version had all controls in one side panel, making the config experience painful and convoluted
2
Scalable Version
Flows divided by context, with config gallery and allowing for scalability
2
Scalable Version
Flows divided by context, with config gallery and allowing for scalability
2
Scalable Version
Flows divided by context, with config gallery and allowing for scalability
3
Presets and Prompt Integration
User can edit image in real time, define presets and remix images via prompts
3
Presets and Prompt Integration
User can edit image in real time, define presets and remix images via prompts
3
Presets and Prompt Integration
User can edit image in real time, define presets and remix images via prompts
TRADE-OFFS
TRADE-OFFS
TRADE-OFFS
1. Automation vs Creative Control
Moving from manual asset dependency to AI-assisted background generation required balancing automation with user autonomy.
Decision
Why
Risk
Mitigation
Removed manual mask upload requirement
Reduce friction & technical dependency
Loss of precision control
AI-generated perspective & foreground adjustment
Introduced pre-built background library
Improve speed & reliability
Visual sameness across dealers
Allow custom uploads & remix capability
Enabled AI-generated backgrounds via prompt
Expand creative flexibility
Unpredictable visual outputs
Real-time preview before final render
Added remix functionality
Encourage iteration
Asset sprawl
Presets + save & organize system
1. Automation vs Creative Control
Moving from manual asset dependency to AI-assisted background generation required balancing automation with user autonomy.
Decision
Why
Risk
Mitigation
Removed manual mask upload requirement
Reduce friction & technical dependency
Loss of precision control
AI-generated perspective & foreground adjustment
Introduced pre-built background library
Improve speed & reliability
Visual sameness across dealers
Allow custom uploads & remix capability
Enabled AI-generated backgrounds via prompt
Expand creative flexibility
Unpredictable visual outputs
Real-time preview before final render
Added remix functionality
Encourage iteration
Asset sprawl
Presets + save & organize system
1. Automation vs Creative Control
Moving from manual asset dependency to AI-assisted background generation required balancing automation with user autonomy.
Decision
Why
Risk
Mitigation
Removed manual mask upload requirement
Reduce friction & technical dependency
Loss of precision control
AI-generated perspective & foreground adjustment
Introduced pre-built background library
Improve speed & reliability
Visual sameness across dealers
Allow custom uploads & remix capability
Enabled AI-generated backgrounds via prompt
Expand creative flexibility
Unpredictable visual outputs
Real-time preview before final render
Added remix functionality
Encourage iteration
Asset sprawl
Presets + save & organize system
2. Flexibility vs Governance (Config Application)
Increasing granularity in configuration targeting introduced scalability — but also governance risks.
Decision
Why
Risk
Mitigation
Moved from single-config-for-all to scoped configs
Enable VIN-level control
Configuration sprawl
Filter-based segmentation
Allowed manual VIN selection
Support edge cases
Human error
Clear visual VIN list & removable selection chips
Added rule-based filters
Reduce unintended application
Harder mental model
Visual mapping of applied scope
Introduced preset saving & sharing
Improve reuse across teams
Inconsistent brand application
Centralized preset management
2. Flexibility vs Governance (Config Application)
Increasing granularity in configuration targeting introduced scalability — but also governance risks.
Decision
Why
Risk
Mitigation
Moved from single-config-for-all to scoped configs
Enable VIN-level control
Configuration sprawl
Filter-based segmentation
Allowed manual VIN selection
Support edge cases
Human error
Clear visual VIN list & removable selection chips
Added rule-based filters
Reduce unintended application
Harder mental model
Visual mapping of applied scope
Introduced preset saving & sharing
Improve reuse across teams
Inconsistent brand application
Centralized preset management
2. Flexibility vs Governance (Config Application)
Increasing granularity in configuration targeting introduced scalability — but also governance risks.
Decision
Why
Risk
Mitigation
Moved from single-config-for-all to scoped configs
Enable VIN-level control
Configuration sprawl
Filter-based segmentation
Allowed manual VIN selection
Support edge cases
Human error
Clear visual VIN list & removable selection chips
Added rule-based filters
Reduce unintended application
Harder mental model
Visual mapping of applied scope
Introduced preset saving & sharing
Improve reuse across teams
Inconsistent brand application
Centralized preset management
3. Power vs Usability (Editing Experience)
Making the system more powerful required rethinking interaction to prevent cognitive overload.
Decision
Why
Risk
Mitigation
Replaced numeric input positioning with drag-and-resize
Improve intuitiveness
Less granular precision
Maintain numeric inputs as optional advanced control
Enabled real-time preview
Reduce trial-and-error rendering
Performance load
Optimized preview rendering before final output
Removed render-first editing model
Shorten feedback loop
Overconfidence in preview accuracy
Clear distinction between preview and final render
Added visual state indicators (active, rendering, applied)
Improve clarity
UI density
Hierarchical layout + controlled spacing
3. Power vs Usability (Editing Experience)
Making the system more powerful required rethinking interaction to prevent cognitive overload.
Decision
Why
Risk
Mitigation
Replaced numeric input positioning with drag-and-resize
Improve intuitiveness
Less granular precision
Maintain numeric inputs as optional advanced control
Enabled real-time preview
Reduce trial-and-error rendering
Performance load
Optimized preview rendering before final output
Removed render-first editing model
Shorten feedback loop
Overconfidence in preview accuracy
Clear distinction between preview and final render
Added visual state indicators (active, rendering, applied)
Improve clarity
UI density
Hierarchical layout + controlled spacing
3. Power vs Usability (Editing Experience)
Making the system more powerful required rethinking interaction to prevent cognitive overload.
Decision
Why
Risk
Mitigation
Replaced numeric input positioning with drag-and-resize
Improve intuitiveness
Less granular precision
Maintain numeric inputs as optional advanced control
Enabled real-time preview
Reduce trial-and-error rendering
Performance load
Optimized preview rendering before final output
Removed render-first editing model
Shorten feedback loop
Overconfidence in preview accuracy
Clear distinction between preview and final render
Added visual state indicators (active, rendering, applied)
Improve clarity
UI density
Hierarchical layout + controlled spacing
Old Version
Old Version
New Version
New Version
HIGHLIGHTS
HIGHLIGHTS
HIGHLIGHTS
Image Editing
A dedicated workspace for composing the final vehicle image before rendering. Users can select from curated background presets, remix or generate new environments, and precisely position and scale the car in real time. The goal was to shorten the feedback loop and give users full visual control before committing to the final output.
Naming and Filtering
Configurations can be clearly named and scoped to specific VINs using filters or manual selection. This ensures long-term manageability as inventory scales, while preventing unintended application across vehicles. Granular targeting transforms configurations from global settings into controlled, intentional actions.
Config Gallery
A centralized view of all configurations, their status, and the VIN segments they affect. Users can duplicate, customize, or remove configs as needed; promoting reuse while maintaining clarity and oversight. The gallery reinforces visibility and governance across the system.
VIN Details
A detailed view of each vehicle and the configuration currently applied to it. From this page, users can trace configuration logic and navigate directly to the source config if adjustments are required. This creates transparency at the vehicle level and reduces ambiguity when managing large inventories.
Image Editing
A dedicated workspace for composing the final vehicle image before rendering. Users can select from curated background presets, remix or generate new environments, and precisely position and scale the car in real time. The goal was to shorten the feedback loop and give users full visual control before committing to the final output.
Naming and Filtering
Configurations can be clearly named and scoped to specific VINs using filters or manual selection. This ensures long-term manageability as inventory scales, while preventing unintended application across vehicles. Granular targeting transforms configurations from global settings into controlled, intentional actions.
Config Gallery
A centralized view of all configurations, their status, and the VIN segments they affect. Users can duplicate, customize, or remove configs as needed; promoting reuse while maintaining clarity and oversight. The gallery reinforces visibility and governance across the system.
VIN Details
A detailed view of each vehicle and the configuration currently applied to it. From this page, users can trace configuration logic and navigate directly to the source config if adjustments are required. This creates transparency at the vehicle level and reduces ambiguity when managing large inventories.
Image Editing
A dedicated workspace for composing the final vehicle image before rendering. Users can select from curated background presets, remix or generate new environments, and precisely position and scale the car in real time. The goal was to shorten the feedback loop and give users full visual control before committing to the final output.
Naming and Filtering
Configurations can be clearly named and scoped to specific VINs using filters or manual selection. This ensures long-term manageability as inventory scales, while preventing unintended application across vehicles. Granular targeting transforms configurations from global settings into controlled, intentional actions.
Config Gallery
A centralized view of all configurations, their status, and the VIN segments they affect. Users can duplicate, customize, or remove configs as needed; promoting reuse while maintaining clarity and oversight. The gallery reinforces visibility and governance across the system.
VIN Details
A detailed view of each vehicle and the configuration currently applied to it. From this page, users can trace configuration logic and navigate directly to the source config if adjustments are required. This creates transparency at the vehicle level and reduces ambiguity when managing large inventories.
USER TESTING
USER TESTING
TESTS
The tests validated not only usability improvements, but the structural soundness of the configuration model at scale.
Power users were able to execute scale tasks, with no friction
Tests indicated smaller issues for novice users, regarding Vehicles page navigation, which was solved for final version
Guardrailing was essential to prevent novice users from interfering in global configurations for auto images. So I needed to isolate the "Config" control in a specific screen, for Admins
The tests validated not only usability improvements, but the structural soundness of the configuration model at scale.
Power users were able to execute scale tasks, with no friction
Tests indicated smaller issues for novice users, regarding Vehicles page navigation, which was solved for final version
Guardrailing was essential to prevent novice users from interfering in global configurations for auto images. So I needed to isolate the "Config" control in a specific screen, for Admins
RESPONSIBILITIES
RESPONSIBILITIES
What did I actually do?
Help clarify where the frictions were, and which problems should be solved
Help clarify where the frictions were, and which problems should be solved
Aligned the team around the right problem to solve
Aligned the team around the right problem to solve
Turned complex product questions into clear decisions
Turned complex product questions into clear decisions
Connected research insights with practical product direction
Connected research insights with practical product direction
Found edge cased and different mental models by testing with users
Found edge cased and different mental models by testing with users
Shipped
Shipped
LEARNINGS
LEARNINGS
LEARNINGS
Enterprise complexity cannot be removed. It must be shaped.
Enterprise systems naturally accumulate complexity as they scale across markets, users, and edge cases. Oversimplifying them only displaces that complexity into manual workarounds and hidden risks. The real challenge is structuring complexity so it becomes understandable, governable, and intentional.
Flexibility without constraints creates operational chaos.
Flexibility at scale is powerful — but without structure, it leads to configuration sprawl and inconsistent behavior. Guardrails, scoped rules, and permissions ensure autonomy doesn’t compromise reliability. Sustainable flexibility always depends on clear boundaries.
Flexibility without constraints creates operational chaos.
Flexibility at scale is powerful — but without structure, it leads to configuration sprawl and inconsistent behavior. Guardrails, scoped rules, and permissions ensure autonomy doesn’t compromise reliability. Sustainable flexibility always depends on clear boundaries.
The hardest part of scaling systems is preventing accidental power.
As systems grow more powerful, the real risk becomes unintended impact. Small changes can propagate at scale without warning. Designing for growth means introducing previews, boundaries, and friction that prevent silent amplification of errors.
Governance mechanisms must be visible, not hidden.
Permissions and rule logic cannot remain invisible backend decisions. When governance is opaque, users act cautiously or make unintended mistakes. Making control mechanisms explicit transforms governance from restriction into clarity and confidence.
Governance mechanisms must be visible, not hidden.
Permissions and rule logic cannot remain invisible backend decisions. When governance is opaque, users act cautiously or make unintended mistakes. Making control mechanisms explicit transforms governance from restriction into clarity and confidence.
