AI-assisted
image retrieval
tool for virtual care
Overview
AI Product, Collaborative, Healthcare, B2B, Saas
Timeline
Jan 2025 - Aug 2025
(ongoing)
My Role
Product Designer, Contribute to UX Research and lead AI product prototyping
Background
InvisibleCare is a virtual healthcare support service that provides on-demand, remote, and personalized assistance for clients' daily activities.
The service integrates live support workers (PSWs) with technology to deliver real-time support. InvisibleCare employs 17 PSWs and plans to onboard five new support staff every year.
Initial Challenge
Delayed Support Due to Escalation Protocol
Currently, challenging situations are escalated to an Occupational Therapist (OT) for additional support, which may not provide timely assistance for immediate client needs.
Insufficient use of Best Practices Images
Invisible Care's built-in toolbox of best practice images is underutilized by PSWs, despite their proven effectiveness and efficiency.
UX Research Overview
Interviews with PSWs and Stakeholders
By interviewing 4 PSWs and 2 company's leadership, we uncovered the stakeholder needs and challenges in current working mode.
PSW Shadowing Sessions
PSW shadowing enables us to observe the challenges, hesitations and dilemmas in PSW's daily work, leading to valuable questions and deeper insights during interviews.
Role-play Workshop
We designed and ran a 24-person (divided into 6 groups) role-play workshop to mimic the thoughts and barriers when PSW is supporting clients.
User Interview & User Shadowing
Based on 4 interviews with PSWs, 4 empathy maps were built and each includes Say, Do, Think, Feel, Needs, Insights to comprehensively understand PSWs' needs and motivations
To further summarize findings, a user persona was built to anticipate user profile and needs, which guides the design and play as a reference for later iterations.
Emily Zhang
-
Age: 24
-
Role: Personal Support Worker
-
Experience: Recently graduated
-
Location: Mississauga, ON (working remotely)
-
Education: Diploma in Personal Support Work
Background
Emily recently completed her PSW program and joined a virtual care company providing support for clients recovering from brain injuries, mental health conditions, and physical challenges. As a digital native, she’s comfortable with technology but still navigating how to balance empathy and efficiency in remote care.
Goals
-
Provide timely, compassionate, and effective virtual support to diverse clients
-
Quickly learn each client's history, preferences, and needs
-
Build rapport and trust with clients in a virtual environment
-
Use digital tools effectively without being overwhelmed
-
Develop confidence in crisis handling and behavior management
Frustrations
-
Searching for appropriate images in the toolbox is time-consuming and not intuitive
-
Information is scattered across multiple platforms, increasing the risk of missing updates
-
Clients often have to repeat themselves due to poor handoff notes or lack of continuity
-
Tone is hard to read in text, making it difficult to respond appropriately
Role-play Workshop
Workshop Design
This group study workshop featured two key activities: a dialogue-based role-play game simulating real-world decision-making scenarios in Personal Support Workers' (PSWs) daily tasks, and a usability test exploring early-stage concepts for navigating best-practice images. The workshop was proposed and designed in close collaboration with the Product Manager to ensure both ethical integrity and methodological soundness.
Research Outcome
Validate previous findings:
-
It is essential to know the context and client info
-
PSWs wish to give personalized support instead of general comforting
New insights:
-
Limitations in virtual care compared to in-person communication
-
PSW's emotional fluctuation due to client's behavior
Feedback on image retrieval:
-
Images might cause misunderstanding
-
Good to have both text and images for clarity and empathy
Stakeholder Analysis
Reframed Challenge & KPIs
PSWs have developed a habitual reliance on OTs for confirmation, even when they possess adequate knowledge and are proceeding correctly, inadvertently increasing the workload for OTs.
Reduce average response time to 1 minute
Lower escalation rate to below 10%
Enable PSWs to review client info within 30 seconds
Brainstorming and Bundling
Semantic Mapping Image Retrieval Model Design
This model is designed to simulate the reasoning and decision-making process of Personal Support Workers (PSWs) in their work. Rather than relying on PSWs to manually extract keywords and connect them to prior knowledge or appropriate resources, the system automates this process. By clearly defining tasks and workflows for the AI, it reduces cognitive load and minimizes the influence of subjective judgment. As a result, PSWs can deliver more efficient, consistent, and scalable support to clients.
AI feasibility research
Test AI agent for image retrieval
Feasibility Documentation
In addition to developing the AI agent to test the feasibility of image retrieval, we also prepared a comprehensive feasibility report analyzing the pros and cons of various AI models, outlining implementation plans, estimating ROI, and addressing potential risks.
Mininum Viable Product (MVP)
Tune Key Words
Recognize Key Words
Associate Key Words
This MVP communicates key design decisions and user workflows to stakeholders, serving as a tool to align the design direction with both business goals and user needs. Building on the proposed Semantic Image Retrieval Model, additional features such as Keyword Tuning, were incorporated to ensure logical consistency and enhance the accuracy of automated responses.
Retrieve Images and Recommending Text
Send Optimized Message
Feedback & Product Scaling to Saas
Feedback from Stakeholders
-
Streamline the manual selection and remove distraction info
-
Add more flexibility to text optimization
-
Consider a more general tool with impact to virtural care industry
Feedback Incorporation and Product Scaling
Based on stakeholder feedback, the "Image Sent/Haven’t Sent" indicator was removed to reduce visual clutter and minimize distractions during use. Additionally, a text alternative was introduced to provide users with two flexible response options.
As a key turning point, we consulted with the Product Manager and identified an opportunity to evolve the tool into a SaaS solution, primarily in the form of a browser extension. This shift enables the product to offer broader, more generalizable support, extending its value to a wider range of businesses and end users.
Ongoing usability testing and iterations ...
