Hello, I have updated my GSoC 2026 proposal for OpenMRS accordingly to what the mentors have suggested me. I am deeply interested in contributing to OpenMRS through GSoC. I have my very own GitHub repo related to disease-prediction using mathematical tools and patient & doctor data management with features of simple chatbot GitHub - aliviahossain/Disease-prediction: A probability calculator using Baye's Theorem to estimate survival chances of a disease based on past hospital data. . I want to integrate some of these features to OpenMRS.
• Lightweight Bayesian Engine (Offline-First): I will stick to Bayes Theorem because it is mathematically ‘light.’ Unlike heavy NLP models, Bayesian calculations are simple multiplications that can run entirely client-side (JavaScript). This ensures the tool is 100% offline-capable and runs on basic hardware without a backend server. • Structured Data over NLP: To avoid the issues of typos, shorthand, and heavy processing in SOAP notes, the tool will pull from OpenMRS Coded Concepts and structured observations. By using standardized data already in the system, we eliminate the need for complex NLP and ensure the input is clinically valid. • Vetted Local Epidemiology (The ‘Data Correctness’ Fix): To solve the issue of inaccurate probabilities, the system will not use a ‘generic’ trained model. Instead, it will use localized probability tables (stored in simple JSON files) based on regional clinical guidelines or data. This ensures the output reflects local epidemiological reality. • Chatbot as a ‘Safety Net’ Interface: The chatbot will act as a UI for these calculations. Instead of a ‘black box’ diagnosis, it will read patient data and provide: Clinical Summaries: Quick digest of recent vitals and symptoms. Risk Triaging: Flagging ‘High Risk’ cases (e.g., Sepsis or Malaria) based on the Bayesian logic and local protocols. Decision Support: Suggesting the top 3 differential diagnoses as a reference for the clinician to verify. This approach transforms the project into a Clinical Decision Support (CDS) utility that is mathematically rigorous but functionally simple enough for the field.
Also, we could eliminate the bayesian screening for now and only implement the chat-bot feature, if it feels too heavy still.
I will be humble enough to say I truly need guidance for it as I am willing to learn and experience through my first GSoC contribution. Also feel free to provide your suggestions on how I should edit my implementation scope and how and what output is expected- anything you want to add or subtract from my proposal.
Thanking You - Alivia Hossain / GitHub: aliviahossain . LinkedIn: https://www.linkedin.com/in/alivia-hossain-513a3a365/