I recently had an interesting experience that I thought might be useful to share with the DHIS2 community, particularly for those working on configuration, maintenance, and deployment of DHIS2 programs.
I needed to update a digitized DHIS2 data collection tool at very short notice. Under normal circumstances, I estimated the work would take 2–3 days of development, testing, troubleshooting, and deployment before the data collection team could use the updated tool.
The challenge was that this wasn’t simply an update to an existing form.
I had originally digitized this particular extraction during the program pilot. The program had now moved into a different stage, and the requirements had changed significantly:
- Some data elements were no longer needed.
- New data elements were required.
- New program rules and validations were needed.
- Existing configurations needed to be modified.
- Other dependencies also had to be reviewed.
So, I decided to try a different approach.
I had access to Claude Code, and with the help of ChatGPT, I converted my requirements and thought process into structured prompts for Claude Code.
1. Map before making changes
The first thing I asked Claude Code to do was not to modify anything.
Instead, I gave it the current DHIS2 JSON configuration and the new paper-based data collection form and asked it to perform a detailed comparison.
The objective was to identify:
- What already existed in DHIS2
- What was missing
- What was no longer required
- Which program rules were missing
- What needed to be updated
- What needed to be created
- Potential conflicts or dependencies
I wanted the AI to first demonstrate that it understood the requirements before allowing it to make any changes.
This turned out to be one of the most important parts of the process.
2. Test before production
I had a test DHIS2 instance available, so I used it as my safety net.
I first tested the proposed changes there rather than immediately modifying the live configuration.
This allowed me to verify that the generated JSON and configuration changes were being executed correctly and that existing parts of the program were not being unintentionally affected.
3. Keep the human in the loop
After the initial testing, I worked iteratively with Claude Code.
Before each major execution, I required a detailed report showing:
- What would be updated
- What would be created
- What would be removed
- Relevant UIDs
- Program rules and program rule actions affected
- Potential dependencies or conflicts
Only after reviewing the proposed changes would I proceed with the execution.
The workflow became:
Understand → Map → Report → Test → Execute → Validate
Rather than:
Prompt → Execute → Hope for the best. 
The result
The task that I expected to take 2–3 days was completed in approximately 6 hours.
That’s a significant reduction in development time.
However, the biggest takeaway for me wasn’t just the time saved.
It was seeing how AI can augment existing DHIS2 expertise.
Claude Code didn’t replace my understanding of the DHIS2 data model, program rules, program rule actions, data elements, validation requirements, JSON structure, or the consequences of making configuration changes.
Instead, it helped me move much faster from:
Requirements → Analysis → Configuration → Testing → Implementation
The combination of:
DHIS2 expertise + ChatGPT + Claude Code + a test environment + human review
created a surprisingly effective workflow.
My biggest lesson
I don’t think the value of AI in DHIS2 is simply giving an AI tool access to your configuration and asking it to “fix” things.
The real value comes from how you structure the work.
Give it context.
Ask it to analyse before modifying.
Ask it to explain what it intends to change.
Require a mapping before execution.
Test in a non-production environment.
Review the changes.
Then execute.
AI didn’t replace the DHIS2 developer in this process. It gave the developer leverage.
I’m curious to hear from the DHIS2 community:
Has anyone else started using Claude Code, ChatGPT, GitHub Copilot, or other AI coding assistants for DHIS2 configuration and maintenance?
If so, what has your experience been—particularly with Program Rules, metadata JSON, SQL, API integrations, or DHIS2 custom development?
dhis2 #DHIS2Community #DigitalHealth #ClaudeCode #ChatGPT ai #HealthInformatics #DigitalHealthSystems #DataManagement #ProgramRules #SoftwareDevelopment