OpenScribe is an open-source AI scribe that records patient encounters and generates structured clinical notes automatically. You keep full control over data, workflows, and patient privacy with no vendor lock-in.

Website for the JuliaHealth organization (powered by Franklin.jl).

Survival functions for DataSHIELD. Package for building survival models, Cox proportional hazards models and Cox regression models in DataSHIELD.

Python SDK for OMOP/OHDSI vocabularies - query 10M+ medical concepts across SNOMED, ICD-10, RxNorm, LOINC & 90+ terminologies via simple API

R SDK for OMOP/OHDSI vocabularies. Query 10M+ medical concepts across SNOMED, ICD-10, RxNorm, LOINC & 90+ terminologies via simple API

Single-file PyTorch pipeline for pathogen class prediction on MIMIC-III/IV EHR data. Hybrid Conv1D+BiLSTM+numeric model, streaming CSV ETL without pandas, ECE calibration, ROC/PR-AUC, and subgroup bias checks. Research and education only.

Detect anomalies in pharmacological profiles using machine learning

This is a package with miscellaneous tools, a set of computational tools and a graphical user interface (GUI) for privacy preserving federated survival model meta-analysis using the DataSHIELD platform. The primary function is to enable data and model documentation.

Benchmark and paper code for hallucination in LLM-based clinical concept normalization (ICD-10-CM, LOINC, RxNorm).

Responsive HTML/CSS/JS template that mimics Sectra PACS Info Panel styling, letting radiology teams embed division-specific quick links (RIS, protocols, teaching files, guidelines, etc.) directly inside PACS.

A curated list of awesome OHDSI tools, packages, and resources for OMOP CDM

A Julia-based machine learning framework to identify predictive biomarkers for chronic Hepatitis B progression from sparse clinical data.

Python library to work with HPO (Human Phenotype Ontology) terms and their gene/disease associations

Quantum kernel SVM for rapid Ebola strain triage (Bundibugyo · Zaire · Sudan · Non-HF). ZZFeatureMap encoding of 20 clinical features. Motivated by the 2026 DRC/Uganda Bundibugyo PHEIC. PennyLane · scikit-learn · imbalanced learning.

Production-grade ML system predicting ICU 30-day readmission risk (MIMIC-IV, 48K patients). Complete pipeline: data cleaning → feature engineering (181 features) → LightGBM (0.79 AUC) → deployed Streamlit app. Healthcare AI portfolio project.

About me (Soumya Banerjee)

Zero-shot clinical note triage and symptom extraction with NegEx-style negation handling, confidence scoring, and Streamlit deployment.

CODEHR: Clinical Open Dashboard for Electronic Health Records

A domain world model for human function, combining ontology, interpretable reasoning, safety constraints, and auditable decision support.

Curriculum Vitae