21 October 2025 Call
French interpretation available
Recap & Updates from Last Meeting
Topic: Exploring AI to Enhance CHT Deployments and Workflows
This community meetup featured pioneering initiatives exploring the use of Artificial Intelligence (AI) within the CHT ecosystem. The discussion highlighted the potential of AI to accelerate development, improve clinical decision support, and maintain the CHT’s core principles of being offline-first and privacy-focused.
1. CHT Multi-Agent System for Code Generation
Overview:
@hareet introduced an advanced concept for a CHT Multi-Agent System designed to move beyond single-query AI interactions. This system would use four specialized AI agents working in concert to generate high-quality, usable CHT code.
-
Research Agent: Understands previous CHT solutions, features, and validated workflows by analyzing GitHub issues and community knowledge.
-
Context Agent: Builds and maintains a growing database of what works, including pitfalls to avoid, becoming smarter over time.
-
Dev Agent: Writes the actual code, adhering to CHT code styles and best practices.
-
QA Agent: Focuses on testing, providing feedback, and ensuring code quality.
Key Benefits:
-
Standardization: Provides a community-standardized AI context and toolset, preventing individual developers from having to configure their own from scratch.
-
Flexibility: Designed to be model-agnostic, allowing the use of any LLM API (e.g., ChatGPT, Claude). This supports future efforts for data privacy and localization, including potential offline models.
-
Ecosystem Integration: The system can interact with other AI tools in the ecosystem, like the Kapa.ai docs assistant, to continuously improve its context and usefulness.
Next Steps & How to Contribute:
A forum post will be published with more detailed designs and subtasks. Community members, especially developers, are encouraged to join the upcoming working session to help divide the work and move the project forward.
Interested? Look out for the forum post and express your interest to get involved.
2. AI for Clinical Decision Support
Overview:
This project focuses on using AI to generate high-fidelity clinical applications that adhere to medical protocols. The core idea is to use AI during the setup phase to generate decision tables and forms, which are then reviewed by medical professionals and run as standard, reliable code on-device.
The Process:
-
An LLM reviews a clinical manual (e.g., from WHO or a national ministry).
-
It identifies gaps and edge cases (e.g., stock-outs) for medical directors to clarify.
-
The AI breaks down the manual and generates Decision Model and Notation (DMN) tables—a format doctors can easily review and approve.
-
These approved tables are compiled into XLSForms for deployment.
-
The system includes extensive testing with simulated patients to validate outputs.
Key Benefits:
-
Safety & Auditability: Avoids LLM “hallucinations” at the point of care by using pre-generated, doctor-approved logic. Every decision is traceable to a source in the manual.
-
Rapid Updates: Allows for quick adaptation to new guidelines (e.g., during a pandemic) by generating new forms for review in days, not months.
-
Leverages Existing Tech: Runs on existing CHT infrastructure as standard XLSForms, making it compatible and offline-first.
Call for Collaboration:
-
Provide feedback from medical and technical staff on the approach and review artifacts.
-
Share existing hardcoded guidelines to use as a gold standard for validating the AI-generated output.
-
Participate in field testing and impact studies with Community Health Workers.
Interested? Please reach out to Professor David @David_I_Levine directly via the forum to explore collaboration. Here is a link to forum post.
Other Highlights & Resources
-
Ethical Considerations: The group discussed key factors for AI in CHT, including data privacy, security, cost, connectivity (offline capability), and accuracy.
-
Offline AI Example: A project was mentioned, which uses an offline AI model to analyze photos of rapid diagnostic tests (RDTs) and provide a result, demonstrating a practical, non-LLM application of on-device AI.
-
Existing AI Tool: The community was reminded of the Kapa.ai AI Assistant already integrated into the CHT docs site, which can answer questions in multiple languages.
Action Items & Next Steps
-
Multi-Agent System: Look out for the upcoming forum post to join the working group.
-
CHT-HW Project: Reach out to @David_I_Levine if your organization can provide medical/technical feedback or is interested in piloting.
A huge thank you for the insightful presentations, and to all community members for the engaging discussion. Let’s continue to explore how we can harness AI responsibly to empower health workers and strengthen health systems.