This is a joint project with Arkus AI. The project involves building and being ready to deploy an AI agent for health and fitness coaching. The development shall be done using Arkus’s AI agent building platform.

Background & Motivation

Effective diabetes management requires sustained behaviour change and strong family support. Traditional apps focus on glucose tracking or diet logging but fail to address the human side—motivation, communication, and emotional connection between patients and caregivers. Recent advances in Large Language Models (LLMs) make it possible to create conversational AI coaches that can support patients holistically, offering education, encouragement, and goal tracking. This project aims to design and prototype such an agent, exploring how it can empower both diabetic patients and their families to collaborate in day-to-day management and long-term behaviour change.

Objectives

  1. Design an AI coaching agent that engages diabetic patients and their families in shared self-management.
  2. Prototype the agent using a modern agent-building platform (e.g., Arkus AI Agent Builder).
  3. Incorporate personalisation, empathy, and family-aware interactions into the dialogue design.
  4. Evaluate usability and perceived usefulness through a small pilot study.
  5. Reflect on ethical, clinical, and design implications of deploying conversational AI in chronic care.

Research Questions

  1. What conversational and behavioural strategies help an AI agent effectively coach diabetes patients and families?
  2. How can personalization (e.g., behaviour stage, relationship dynamics, preferences) improve engagement?
  3. What are the usability and ethical considerations for deploying such an AI in healthcare contexts?

Literature

Conversational agent interventions in diabetes care by M. Shegal et al. (2025) — This is a systematic review evaluating conversational-agent (CA) interventions in diabetes care (effectiveness, acceptability, safety).

Family‐based interventions for adults with type 2 diabetes: A systematic review and meta‑analysis by K. A. Matrook et al. (2025) — This work focuses on the effectiveness of family-based interventions in T2DM.

Deliverables

  1. Functional prototype of the AI coaching agent.
  2. Design documentation (personas, flows, architecture).
  3. Pilot study report and analysis.
  4. Ethical reflection and deployment guidelines.
  5. Final thesis report and presentation

Prerequisites

Strong knowledge of PyTorch and Large Language Models is required. We also expect you to be already familiar with building agents using LLMs and deploying them in the real world.

How to apply

If you are interested, send:

  • A short statement of interest (1 page max): What are your interests? How are you going to apply them to this project?
  • Your CV
  • Relevant transcripts

Deadline to register your interest: 10th of November

Contact

Contact: