In response to Singapore’s aging population and the rising prevalence of chronic conditions, the Singaporean government has shifted focus from curative care to preventative care. As part of national efforts to promote healthier lifestyles, PSYKHE was assigned to assess and test the feasibility of GenAI solutions for delivering personalised healthcare advice and ultimately, increase adherence to clinical prescriptions.

‍

In collaboration with technology and healthcare providers, PSYKHE explored the potential of using GenAI to offer:

• AI generated diet and exercise recommendations
• Hyper-personalised prescriptions based on user profiles
• General diet & exercise inquiry

‍

Over 14 weeks, PSYKHE collaborated with technology and healthcare partners to envision, design and test the capability of GenAI to provide hyper-personalised health recommendations. The pillars of our approach consisted of:

• Collaboration: Co-creation between numerous healthcare, transformation and promotion agencies
• Technical feasibility: Continuous evaluation and refinement of prototype build to push the limits of generative AI in a safe setting
• Comprehensive testing: Parallel quantitative testing with clinicians and residents to evaluate appropriateness and usefulness of our proposed solution

Our approach followed 6 design sprints across 3 key stages from discovery to assessment.

‍

Stage 1: Discovery & Define

In collaboration with healthcare agencies, the team defined the initial requirements of what “good looks like” for a proof of concept GenAI tool.

Through collaborative workshops involving multiple technical partners and healthcare agencies, we developed the initial requirements for a resident-facing self-help tool that would provide personalised diet & exercise recommendations on the fly. The proof of concept centered on 3 key defining features:

• Resident facing: To implement a GenAI tool that residents would use to take ownership and tailor their own diet & exercise prescriptions.
• Personalised: To allow for the personalisation of prescriptions based on resident preferences and health statuses
• Appropriate: To implement safety measures that would ensure AI-generated recommendations were safe and appropriate for residents.

These workshops also helped to foster a sense of ownership and stakeholder buy-in over the concept.

‍

Stage 2: Build & Test

With the concept and requirements in place, PSYKHE collaborated with technology providers to build a functional web prototype that could be tested across technical feasibility and user evaluation.

This stage involved frequent collaboration across multiple build streams:

• Data stream: The collation, vetting and training of our LLM model using appropriate healthcare data sources
• Backend stream: The development of the backend infrastructure and API builds for proof of concept
• Frontend stream: The design, development, testing and refinement of the user facing experience.

‍

Stage 3: Assessment & Wrap-up

After numerous refinements and iterations of our proof of concept build, the team delivered a fully functional web-prototype that allowed users to:

• Receive personalised diet & exercise plans tailored to their health conditions
• Tailor their health prescriptions based on their individual preferences
• Inquire about general diet & exercise through an interactive chatbot

As part of the final assessment, the team conducted user testing validations across 20+ residents and clinicians - alongside technical testing of 6 technical domains.

‍

‍