/
2026-01-29 AI Meeting Notes

2026-01-29 AI Meeting Notes

Attendees: Paul Biondich, Piotr Wojciech Mankowski, Jan Flowers, Ian Bacher, Veronica Muthee

Project Alignment

Foundation for AI-enabled feature

Three-Part Technical Approach

  • Part-1: Improvements to the OMRS core

    • Enhancing core platform capabilities to support AI integration

  • Part-2: CQRS Pattern 

    • Separate data models for writes vs. reads

    • The current EAV (Entity-Attribute-Value) model is difficult for AI to interpret

    • Solution: Flatten data structures for AI consumption

  • Part 3: Data Flattening & AI-Ready Architecture

    • The current EAV model is difficult for AI to read - needs flattening

    • Current Elasticsearch usage is minimal (patient names, identifiers, concept dictionary)

    • Need to flatten concepts and observations

    • Create person-level flattened data (not just surface-level)

IPS-Like Patient Object

  • Create a standardized, flattened patient record that AI can easily consume

  • Similar to the International Patient Summary (IPS) format

  • Google AI expect FHIR-formatted data? Should we use that? 

  • Remove verbosity from current data structures ~ make it simple

  • Leverage FHIR object model conventions without full FHIR implementation overhead ~ use parts that help and not the entire

Key Technical/Implementation Considerations

  • Leveraging existing Elastic Search or use of alternative solution? 

  • Create a persistence layer optimized for reads (separate from writes)?  

    • Special copy of data that's super fast to look at? 

  • Build hooks that allow AI systems to view/access flattened data /so that AI can easily peek at simplified data

  • Flatten at the concept level for better AI interpretation?

  • FHIR resources considerations ~ need to determine how observations fit

  • Consider custom documents vs. standard FHIR resources ???

  • Integrating LLM into OpenMRS requires trade-offs?

  • Prototype development with user testing

  • Create the first version of the read-optimized data layer

  • Not an appendage solely for AI - building for multiple use cases

  • Community consensus building through Talk post

Team and Resources

  • Rafal

    • Engagements at the moment: 1) Privilege segregated data access (based on location ~ DRC use case, where data is hidden/shown based on where the user is accessing from) 2) Integration Middleware Development, i.e., design of secure interoperability solutions that will enable safe patient data exchange between OpenMRS and external healthcare systems, such as laboratories

  • Ian Bacher

  • Piotr Mankowski

  • Graham and Jonathan Taish

  • Partnership outreach needed

  • Coordination with Beryl for potential partnership outreach

  • Tendo …. regarding OpenMRS AI initiatives

Timeline

  • 6 Months, January - June 2026

  • Question: Is June realistic for deliverables?

Contract & Funding

  • Contract has no restrictions

  • Charitable donation structure

  • Reports and deliverables? Need to cross-check

Communication & Documentation

  • Establish communication channels

  • Documentation approach to be defined

  • Community engagement through Talk posts

  • Engage EMR4All 

Next Steps

  • Paul/Jan will strategize with Beryl on partnership outreach

  • Paul to meet with Graham to fine-tune the approach

  • Ian to draft a community Talk post about the approach

  • Ian to define Rafal contract…

References