Purpose and objectives

This playbook provides a methodology for applying human-centered design (HCD) to develop locally relevant datasets for training and evaluating medical large language models (LLMs). The objectives of this playbook are to:

• Establish processes for gathering contextually appropriate medical scenarios from frontline health care workers (FLWs).
• Ensure artificial intelligence (AI) benefits reach underrepresented populations
• Address bias in existing LLMs by incorporating diverse medical knowledge
• Create frameworks for ongoing dataset improvement
• Empower local stakeholders to shape AI tools reflecting their needs

How to use this playbook

Use this resource as:

• A guide for creating inclusive medical datasets
• A complement to evaluation frameworks
• A roadmap for replicating methodologies across contexts