22 April 2025 Call
French interpretation available
Topics discussed
- Revisiting a previous topic discussed about AI. @alexkombo has explored AI with the CHT and continued the conversation. Use cases and questions that were considered in the last weeks:
- How do we integrate AI tools with eCHIS, as it is offline first?
- External integration: the AI models are external to eCHIS and the data is consumed via APIs
- Costs are expensive. Some models are more expensive than others
- Risk: vendor locking
- As eCHIS is offline, we can’t get the best of the AI tools as users are offline
- Native integration within the CHT ecosystem on health workers phones
- Use open source AI
- Challenge: AI dev is very different than SW dev. Limited computation power on user device. Limited storage space.
- Dev complexity, slow pace of iteration
- Hybrid: part of it on eCHIS, part of it on the phones
- 2 existing projects at the moment (apps that integrate with the CHT - 2apps)
- Not very scalable, as we cannot have another app installed on the phone for every different use case.
- Some use cases:
- Task prioritization (risk-based) and route optimization for Community Health Promoters (CHPs).
- Enhanced data analysis and insights generation. Users wants to have access to insights about data that make sense for them. AI could help with that. Pull data, do some processing, get insights.
- Event-based surveillance for disease outbreaks or health trends.
- Training and support (e.g. train the CHPs when new features are added, e.g. via a chatbot via WhatsApp). Train the AI on all interactions and support data. Is it a UX problem? Is is a data problem? Is it a feature problem?
- Chatbot on the docs site for asking any question
- Data analytics
- Due to data quality issue that the model could be biased
- It makes more sense to invest on other activities that have more impact on data quality
- Is it possible for AI to help support app developers or code contributors? Can it support make the implementers be more successful?
- Vibe-coding: some developers already use AI extensively when doing app development to the CHT, e.g. form development or testing. Can we get the AI to do that part of development work?
- There are tools out there that ingest code, and generate text that can be ingested into AI tools. https://gitingest.com is one of them. It turns any git repository into a simple text digest of its codebase. This is useful for feeding a codebase into any LLM.
- Some developers used CoPilot & ChatGPT for translating tests when configuring app and it worked quite well. French → ChatGPT → English.
- AI needs a lot of context, and that’s what the biggest challenge. We can’t use real people data for training AI. How can we tackle the problem related to confidentiality?
- Anthropic uses MCP
- Experience of first time contributors
- Good experience, but faced some challenges
- The “Good First Issues” could be more challenging than expected
- The biggest challenge: running the app locally. Some docs are lacking.
- The community is supportive both on GitHub and the forum
- CHT Academy app dev course is very helpful
- Dev can be tougher on macOS
- Need to list skillset needed on Good First Issues and squads