@Madiro in partnership with @OpenConceptLab has made progress using AI to help rapidly set up OpenMRS configuration, like concept mappings, paper form conversion, and translation
@OpenConceptLab has been experimenting with using MCP to make querying the OCL api easier, which in the future could possibly help OpenMRS terminology managers
@Mekom added to Ozone’s roadmap the goal of providing an AI prompt to consume Ozone Analytics flattened data leveraging Meditron-3
Bahmni (since January 2025) added Medispeak Integration, which leverages OpenAI APIs for speech-to-text transcription and context-based form filling.
…and more! Come tell us what you’ve been working on!!
Can’t make it this Monday?Don’t worry -This call will be recorded. Let me know if you are interested in a subsequent call, and then we will set that up
I know people are really focused on LLMs, but over the last month, I have taken an interest in small language models and their potential for specialised use cases in resource-constrained environments. I am thinking of models that can be run on a CPU only and still produce good results .
we at @EMR4All have done some experiment with SLM’s on offline edge devices like PI’s, but the one’s we’ve tried are fast but don’t have enough context, they therefore hallucinate almost 90% of the time**(remember we’re dealing with patient data, any slight mistake can mean a death sentence to the patient),** , that’s why we’re bringing these combinations, agents, models + tools, RAGs, models + mcp e.t.c. , but better callibration is the now the way I guess
Both SLMs run smoothly, LLM’s of a few billion params(1-7), run well, but need just about enough context
But will be happy to hear of your findings with SLM’s
@EMR4All We are adapting the existing English based queries that return results and SQL queries run to a generic OpenMRS Module. The goal is for the existing features to be made available via an OpenMRS module and also for Clinicians to be able to query the EMR for population-level health questions by typing their questions ( e.g. “What percentage of my patients have diabetes?”) into a frontend built-in search bar.
The query will be sent through this module to locally deployed LLMs and the results will show, explaining the derived answer. Citations and the SQL used will be included, so that surprising findings can be confirmed.
The module uses tools and utilities common to the OpenMRS environment, such as the Spring Framework, Hibernate, Liquibase, Slf4j, JUnit, and Mockitoand includes only LangChain4j, Testcontainers and Lombok as additional Libraries.
particular focus on Unit & Integration Testing | Testing LLM Responses to reduce Hallucination.
UserMessage userMessage = UserMessage.from("What is the name of the process by which the body breaks down food?");
ChatResponse response = model.chat(userMessage);
AiMessage aiMessage = response.aiMessage();
assertThat(aiMessage.text()).contains("digestion");
assertThat(aiMessage.toolExecutionRequests()).isEmpty();
ChatResponseMetadata metadata = response.metadata();
assertThat(metadata.modelName()).isEqualTo(MODEL_NAME);
Sample;
If we gave a wrong Answer for instance; the result would be as follows.
[ERROR] Failures:
[ERROR] ExpertSystemServiceTest.chat_should_generate_valid_response:62
Expecting actual:
" The process by which the body breaks down food is called digestion. It involves several steps that break down complex carbohydrates and proteins into simpler, more usable forms. These simpler substances can then be absorbed by the bloodstream to be used as energy or stored in the form of glycogen for later use."
to contain:
"Respiration"
Hi All ,
Given the rapid advancements, AI is becoming a critical tool for Health .In our recent discussions, people have showcased impressive work of AI . However, a key question remains: how does this work benefit the wider community ?
To maximize our collective impact, let’s sue this forum for brainstorming and begin designing a core AI module for OpenMRS ; thing we can start with
Identify priority use cases for AI within OpenMRS (e.g., clinical decision support, predictive analytics, data quality, reporting).
Design AI architecture for starting a Module that will be integrated in OpenMRS
Engage developers to build up the moduel workig together on this common goal
I believe that by starting a focused, community-driven effort to design a core AI module, we can develop tools accessible to strengthen the OpenMRS ecosystem.
This is what we have been up to for the ESM, we are adapting the existing @Emr4all A.I work to OpenMRS, through the weeks we have talked about our progress during the A.I calls…Join the bi-weekly meetings to get more..
Let’s talk this week about AI and OpenMRS
Google Calendar Invite 
Can’t make it this Monday? Don’t worry - This call will be recorded. Let me know if you are interested in a subsequent call, and then we will set that up 
.
