RAG-Based AI Support Tool for Community Health Workers — Sierra Leone
Fully deployed AI support tool for Community Health Workers in Sierra Leone — from field research (April 2025) to live implementation (May 2026)
The Challenge
Community Health Workers (CHWs) in Sierra Leone operate in low-connectivity, low-digital-literacy environments, often without reliable access to trusted clinical guidance during patient interactions. The Ministry of Health Sierra Leone's ambition to transition from paper-based to digital care — across community health and patient management — required tools that work in real-world field conditions, not just controlled environments.
The Approach
Working with Medtronic LABS and the Ministry of Health Sierra Leone, I spent time in April with CHWs in the field to explore a RAG-based (Retrieval-Augmented Generation) AI support tool delivered entirely over WhatsApp — designed for low-connectivity, low-digital-literacy environments. The goal was to observe actual usage patterns, collect usability feedback, and understand how AI-powered tools perform in contexts defined by limited connectivity and varying digital literacy.
The Result
The field research phase surfaced three defining lessons that directly shaped the tool's design. By May 2026, the tool was fully implemented and deployed — giving Community Health Workers in Sierra Leone real-time access to trusted clinical guidance over WhatsApp, at scale. Retrieval quality, low-latency responses, and CHW onboarding were the three pillars that made adoption possible in the field.
Field Lessons
Retrieval only works if what you're retrieving is trusted
A hallucination isn't a UX problem — it's a risk. In healthcare, incorrect AI-generated answers aren't just unhelpful, they're potentially dangerous to patients and health workers alike.
Latency isn't inconvenient — it's the adoption barrier
In low-connectivity environments, a slow response is the difference between a tool that gets used and one that gets abandoned. Performance is a product requirement, not an optimization.
Human connection is still needed
Good understanding of the 'why' is critical to avoid search loops and tool burden. AI tools without proper onboarding and context lead to over-reliance or abandonment — neither is useful in the field.
Deliverables
- Fully implemented RAG-based AI chatbot deployed over WhatsApp
- Field research with CHWs in Sierra Leone (April 2025)
- Usability testing and feedback loops in low-connectivity environments
- Full production deployment serving Community Health Workers (May 2026)
Tools Used
Want similar results for your organization?
Let's discuss how I can help improve your workflows, build scalable systems, and coordinate AI tools that deliver real operational impact.