I’m starting this thread to focus specifically on the technical implementation and extension of the current Chart Search AI Module - ChartSearchAI. This will allow the other thread to remain dedicated to general AI applications in OpenMRS.
Hi everyone,
I am moving the technical discussion for ChartSearchAI here. My goal is to implement the vision shared by @burke and the team for an integrated, AI-supported patient chart search.
I’ll be sharing progress on the logic here. Feedback is highly welcome.
I do have question. I work in mental health in residential treatment so often our residents(patients) come with incredibly long documentation. Imagine, every day staff members of all institution types must write notes on every patient in their care. Extend this by perhaps a team that is made up of a nurse/caregiver, case manager and a clinician (granted clinicians and case managers don’t write daily on all their patients but they read more). And then, multiply this by the amount of time someone has spent in the system. Some their whole life, some only a few years.
Secondly, patients who have been institutionalized for most of their life, you know, prior to the digitization of forms and documentation, come with scanned PDF files of their notes. Thankfully they’re mostly from typewriters and not hand written.
What I want to be able to do, is identify behavioral patterns that show up in the notes i.e
Patient has a history of refusing medications
During times of distress “this” was helpful in returning to baseline behavior
this patient is known to be aggressive and sexual to female staff members
“this” was a significant or traumatic event for the patient
This use case , alongside any security concerns that staff should know, is probably the most important as it helps us situate us to the type of trauma a person has endured.
These mainly come from clinician reports that document significant life events.
Let’s assume I will have access to the best-case target you identified. I understand this OpenEHR is tailored for medical not behavioral health but given the planned architecture, is this something that is feasible? If you have any questions or want me to clarify more on my user story let me know.
@alekee OpenMRS adapts to residential behavioral care. It can be customized for mental health workflows through its concept dictionary, while your local hardware securely digitizes legacy records to identify critical patient patterns.
An automated data ingestion pipeline parser digitizes scanned documents, preserving structural metadata before converting the text into embeddings for semantic search. The OpenMRS frontend module integrates a conversational search interface with a static safety dashboard with the objective summaries, providing verifiable citations directly linked to the original scanned documents. This architecture runs entirely offline on your dual Tesla P40 server, ensuring sensitive patient data remains secure while local language models synthesize the retrieved records.
A vector RAG or vectorless Reasoning-based PageIndex RAG database and the relational database with or without a GraphRAG.
