CareAR
UX/UI Design · ML/AI project · Web & iPad app & mobile app
UX strategy, research, wireframing, UI, design system
Services
Client
Xerox / CareAR (USA)
Timeline
Feb 2022 - Mar 2023
Design team
Senior UX Designer
Junior UX designer
About the client
Context
CareAR, a Xerox company, offers an augmented reality platform for customer, field, and IT support, improving outcomes and experiences. It reduces costly dispatches and speeds resolutions with self-solve content, remote AR guidance, live video collaboration, and system integration, enhancing safety and service workflows.
Project overview
CareAR is a Xerox-backed platform that enables enterprise service teams to deliver intelligent, AR-powered remote assistance. It is enhanced by Alto AI, an intelligence engine developed at Xerox PARC, designed to support real-time visual guidance and automation through machine learning.
As the UX/UI Designer, I was responsible for designing internal tools for AI model creation and data preparation, including a 3D scanning and labeling application and a web-based admin dashboard used by AI. The goal was to simplify and streamline the process of generating high-quality training data for computer vision models. These models were later deployed across the CareAR ecosystem to support intelligent object recognition and contextual guidance.
The system supported:
Image recognition and search
Intelligent text/image retrieval
Device state recognition and dynamic labeling
The project lasted over three years and involved a team of 19 people, including backend, frontend, DevOps, machine learning engineers, UI/UX designers, Business Analyst, and QA specialists. We successfully delivered the tools into production use for internal teams, and finalized the system for integration into the CareAR and Alto AI platform.
Task
Design internal tools to make it easier to create machine learning models.
The process of preparing training data for object recognition was fragmented and time-consuming. Our goal was to create intuitive interfaces for scanning physical objects, applying labels, and assembling datasets — enabling AI teams to generate production-ready models quickly and with minimal manual effort.
Challenge
Translate highly technical AI workflows into intuitive tools for non-technical users.
Creating training datasets typically required manual coordination between AI engineers and designers, with tools for labeling, validation, and model configuration. The main challenge was to bring everything together in one place, while still giving expert users the control they needed.
Design a unified workflow for generating ML models from 3D scans — accessible to both AI teams and non-technical operators.
We developed a modular system that allowed users to scan objects, apply semantic labels, and assemble datasets through guided steps. Expert mode enabled full control for data scientists, while simplified views supported faster onboarding. As a result, the organization significantly reduced the time needed to prepare training data and improved the consistency of model quality across teams.
Solution
My contribution
I designed internal tools that supported AI model training based on labeled 3D scans. This included both a mobile 3D labeling app and a web-based admin dashboard. This included:
A 3D labeling app for scanning objects and tagging components
A training workflow UI for selecting scans and configuring ML experiments
Dashboards to visualize model performance (mAP, confusion matrix)
Integration with Intelligent Search indexes used by Alto AI
I collaborated with ML engineers and the stakeholders' product team to simplify complex processes into intuitive, scalable interfaces.
Research approach:
01 / Domain research
Primary research:
Secondary research
Studied design patterns for ML tools and dashboards
Referred to the official guidelines for creating task-based admin tools
Applied established UX principles to form design logic and hierarchy
Since there was no formal user research process or discovery phase, I conducted self-driven domain research to understand:
The machine learning pipeline used by the client
Typical UX patterns in tools for dataset creation and model training
Common pain points when labeling and managing 3D scan data
This helped me translate the stakeholder's functional requirements into effective user interfaces.
Internal knowledge gathering: reviewed technical documentation and API logic provided by the client
Analyzed the steps in the model training flow based on functional specs and meetings
Worked closely with product owners and engineers to clarify edge cases and success criteria
Business goal
The goal was to deliver internal tools that help AI teams efficiently generate machine learning models from 3D scan data. The platform needed to support the full workflow — from selecting and labeling scans to launching training and publishing models — and ensure consistent UX across teams.
Research goals:
Get familiar with a new domain (AI model training and 3D labeling), which I had no prior experience with.
Understand the user workflows, terminology, and technical constraints.
Apply UX best practices to design clear, scalable, and intuitive tools for expert and semi-technical users.
Since the domain was entirely new to me, I researched how machine learning pipelines work, how labeled 3D scans are used for training, and how model performance is validated — to make informed UX decisions.
02 / UX audit
Performed a UX audit of the existing app and admin tools to identify usability gaps, navigation issues, and areas for improvement
Used the audit results to inform layout decisions, reduce cognitive load, and align designs with user expectations
03 / Wireframing
I mapped out the user flows for scan selection, model configuration, and performance validation before moving to high-fidelity mockups.
CareAR iPad app
CareAR web portal
Alto AI web portal
Alto AI Detected app
04 / UI Design & components
Easy-to-use UI components with a clear layout, making the training flow consistent and flexible for expert users.
Before
After (design system for every product I worked on this project)
Created and scaled product-specific design systems where none existed before, bringing consistency and structure to multiple tools within the CareAR platform.
Design system
05 / Feedback & iterations
Regularly shared designs with cross-functional team members — including ML engineers, product owners, and QA — to gather feedback early and often.
Iterated on UX flows based on technical constraints (e.g. model training limitations, scan metadata structure).
Adjusted components and layouts after dev handoff reviews to ensure clarity and feasibility.
Simplified complex steps (e.g. scan selection, label previews) based on internal testing and stakeholder suggestions.
Introduced small interaction tweaks (like live counters, disabled states, or batch selectors) to improve usability after hands-on testing by team members.
06 / Final Outcome
The designs were delivered to development and are now used by internal CareAR teams. The new flow simplified model setup and improved consistency across labeling and training processes.
My work contributed to the delivery of AI-powered tools that supported the CareAR Assist ecosystem. As part of Xerox’s broader service experience, CareAR has achieved measurable impact:
57,000 fewer site visits
9% increase in remote issue resolution
60% less training time
Over 594,000 kg of CO₂ emissions avoided
34% global adoption rate (2022)
These results reflect the effectiveness of well-integrated AR, AI, and support tools working together to improve service quality and sustainability.
After final testing and review, CareAR project was handed over to the client’s team and released to production. All design assets and documentation were provided to ensure a smooth transition and long-term maintainability.
