CareCompass
Executive Summary
Problem
Caregivers struggled to source trustworthy dementia-care information, increasing stress and burnout.
Role
UX Lead across research, conversational design, service mapping, and AI-assisted experience design.
Approach
Human-library workshops, caregiver interviews, ecosystem mapping, RAG-powered conversational assistant.
CareCompass is a conversational care assistant designed to support caregivers in Singapore who are caring for loved ones with dementia. The platform addresses the critical challenge of sourcing relevant, trustworthy information on day-to-day care duties, which is currently time-consuming and frustrating for caregivers.
My responsibilities included:
• Facilitating research synthesis and problem framing
• Mapping caregiver pain points and service ecosystems
• Defining the conversational experience and user journey
• Designing interaction flows for AI-assisted recommendations
• Translating fragmented care services into structured, understandable pathways
• Balancing trust, usability, and responsible AI considerations
• Leading iterative prototyping and stakeholder feedback loops
Beyond interface design, the role required aligning multiple perspectives within a fast-moving, ambiguous environment — balancing technical feasibility, caregiver needs, and service delivery realities.
As part of the design team, we worked to create a mobile-first conversational assistant that provides personalized recommendations and actionable steps tailored to caregivers' needs.
- The project focused on first-time caregivers of dementia patients who need access to trustworthy, personalized information on dementia-related support and services.
We spoke with various governmental bodies and established a repository of pain points via a human library workshop, and mapped out the caregiving ecosystem and its stakeholders and connections.
Speaking to caregivers personally within our primary research also allowed us to narrow down on perceived issues sounded out by caregivers themselves. Once the repository of pain points were established, we mapped out a value matrix before deciding that the problem space we’re tackling is the optimal one.
Speaking to caregivers personally within our primary research also allowed us to narrow down on perceived issues sounded out by caregivers themselves. Once the repository of pain points were established, we mapped out a value matrix before deciding that the problem space we’re tackling is the optimal one.
At a systems level:
- we engaged governmental bodies and care organisations through workshops and human-library sessions to understand service fragmentation, support pathways, and institutional pain points.
At the user level:
- we conducted primary research with caregivers to understand daily caregiving routines, emotional burden, decision-making challenges, and how they currently search for support information.
The problem was not only information access: caregivers often struggled to understand:
• Which services were relevant to their situation
• Whether sources could be trusted
• What action to take next
This insight reframed our problem from “improving search” toward enabling decision support.
• 8.6 hours spent per day on caregiving duties, creating significant time pressure
• 2 out of 5 caregivers are at the risk of depression due to the burden of care
• 2x higher burden of care on caregivers in Dementia than in other conditions
Specifically, sourcing relevant, trustworthy information on top of day-to-day care duties is time consuming and frustrating.
For first-time caregivers of dementia patients who need access to trustworthy, personalized information on dementia-related support and services, CareCompass+ is a conversational assistant that provides personalized recommendations and actionable steps tailored to caregivers' needs.
We broke the problem down into 3 parts:
During early prototyping, we initially explored improving search and information aggregation. However, synthesis from caregiver interviews revealed that access alone was insufficient: caregivers struggled to decide what action to take next.
This insight led us to pivot toward a guided decision-support model focused on actionable recommendations rather than passive information retrieval.
This insight led us to pivot toward a guided decision-support model focused on actionable recommendations rather than passive information retrieval.
A key design decision was choosing a Retrieval-Augmented Generation (RAG) approach rather than a generic chatbot experience.
Caregiving decisions involve high emotional and practical stakes, meaning trust and accuracy were critical.
Instead of relying purely on generative responses, CareCompass retrieves information from trusted caregiving agencies, government resources, and verified service providers before generating contextual recommendations.
Caregiving decisions involve high emotional and practical stakes, meaning trust and accuracy were critical.
Instead of relying purely on generative responses, CareCompass retrieves information from trusted caregiving agencies, government resources, and verified service providers before generating contextual recommendations.
This allowed us to balance:
• Trustworthiness: grounding answers in verified sources
• Personalisation: tailoring recommendations to caregiver context
• Reduced cognitive load: surfacing relevant next steps instead of overwhelming information
• Transparency: helping users understand where recommendations came from
We intentionally designed the system not as an answer engine, but as a guided decision-support assistant.
• Trustworthiness: grounding answers in verified sources
• Personalisation: tailoring recommendations to caregiver context
• Reduced cognitive load: surfacing relevant next steps instead of overwhelming information
• Transparency: helping users understand where recommendations came from
We intentionally designed the system not as an answer engine, but as a guided decision-support assistant.
We approached this not as building features, but as designing a system that helps caregivers make decisions across fragmented services.
CareCompass differentiates itself through four key value propositions:
CareCompass differentiates itself through four key value propositions:
The caregiving ecosystem contains fragmented information. Surfacing everything increased cognitive load, so we prioritised contextual relevance over comprehensiveness.
As an AI-assisted product, responses needed to feel personalised without compromising trustworthiness. This informed our decision to use a RAG approach grounded in verified agency information.
Operating in an accelerator environment required balancing ambition with an achievable MVP that could demonstrate value quickly.
- Generic information doesn't address specific needs
- Reduces information overload and decision fatigue
- Builds trust through relevant suggestions
- Saves time by filtering options
- Improves outcomes by matching needs to services
Why Conversational Interface?
Lower barrier to entry: No need to navigate complex menus
Natural interaction: Mimics talking to a knowledgeable friend
Accessible: Works for users of varying technical abilities
Flexible: Can handle varied and specific queries
Empathetic: Provides emotional support through the interaction
Why Mobile-First?
Caregivers need information on-the-go
24/7 accessibility for urgent questions
Can be used during care activities
Lower friction than desktop research
Enables voice input for hands-free use
Why Personalized Recommendations?
Generic information doesn't address specific needs
Reduces information overload and decision fatigue
Builds trust through relevant suggestions
Saves time by filtering options
Improves outcomes by matching needs to services
Qualitative Impact:
- Reduced caregiver stress and information overload
- Increased confidence in finding appropriate care services
- Improved accessibility to trustworthy information
- Created foundation for future data-driven policy planning
Strategic Value:
- Established platform for gathering caregiver behavior data
- Established a scalable foundation for future service integration and caregiver support pathways.
- Positioned for scaling across different care conditions
- Demonstrated viability for government partnership
• Became a finalist and winning solution
• Was demonstrated to relevant agencies and stakeholders
• Progressed into a funded accelerator programme
• Continued development over a 3-month period and remains active
This progression allowed us to move beyond conceptual prototyping into understanding how an AI-assisted care model could operate in real-world caregiving contexts.
CareCompass reinforced my belief that impactful AI products are not just about intelligent systems, but about helping people make better decisions in high-friction moments.
The project strengthened my experience in responsible AI, service ecosystems, and translating fragmented information into meaningful user outcomes.
- Expand service provider database
- Implement user accounts for saving recommendations
- Add review and rating system
- Enhance recommendation algorithm with machine learning
- Launch marketplace for service provider acquisition
- Implement data analytics dashboard for government agencies
- Expand to other care conditions beyond dementia
- Add community features for caregiver peer support
- Become the trusted platform for all caregiving needs in Singapore
- Use data insights to inform national caregiving policy
- Create sustainable business model through marketplace
- Scale regionally to other markets with aging populations
Thank you for reading till the end.