A deep learning ensemble model built for multi-label chest X-ray classification in the Grand X-Ray Slam Division B Kaggle competition. The system integrates three pretrained CNN architectures, data augmentation, class-aware training, and staged fine-tuning to achieve competitive performance on thoracic abnormality detection.

AI-powered healthcare web application that predicts the risk of multiple diseases using machine learning models built with Python, Scikit-learn, and Streamlit.

MIEO (Masked Input Encoded Output): self-supervised embeddings for clinical tabular data; handles missing values and mixed types for CVD risk prediction.

Cancer risk factors dataset (2,000×21): lifestyle/environment/genetics + derived Overall_Risk_Score/Risk_Level with leakage note (ML & EDA-ready).

Planning to Fairly Allocate: Probabilistic Fairness in the Restless Bandit Setting (KDD 2023)

Cancer risk classification pipeline: leakage detection, Stratified CV (macro-F1), model benchmarking (LogReg/RF/Calibrated SVM/XGBoost), and interpretability (permutation importance).

Agentic personal medical assistant that reasons over medical data using multi-agent orchestration, with leveraging mutiple ML/DL pre-trained models, with addition to relational and vector databases.

End-to-end MLOps project for colorectal cancer survival prediction using MLflow, DagsHub, Kubeflow, and Kubernetes. Features automated ML pipelines, experiment tracking, and containerized deployment.

Simulated hospital deterioration dataset: 10,000 admissions, hourly vitals+labs over a 72h cohort, with ML-ready tables incl. “next 12h” early-warning target.

Hospital deterioration early-warning EDA (first 72h): cohort KPIs, event timing, risk-score behavior, physiologic trajectories, and a toy alert-policy view.

12h clinical deterioration early-warning baseline: tabular models (LogReg / HistGradientBoosting / optional XGBoost), simple probability ensemble, and cost-based threshold/policy tables exported for dashboards.

Implement logistic regression using Python and scikit-learn to classify malignant vs. benign tumours from the Breast Cancer Wisconsin (Diagnostic) dataset

Data-driven modelling and optimization of Scotland’s A&E system for patient flow, waiting times, and site planning.

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.

Portfolio demonstration: A lightweight demo calling Google Gemini 1.5 Flash from Python to generate healthcare ML content.

Advanced machine learning system for cardiovascular disease risk prediction using clinical biomarkers and patient health indicators

Clinical Reasoning Platform Base Repository

Interpretable ML pipeline for modeling postprandial glucose responses (PPGR) from multimodal signals (CGM + meals + context). Code only—no data.

Clinical + Genomic **RAG evaluation (pro)** with hybrid retrieval (BM25 + embeddings), stronger faithfulness, YAML configs, and HTML dashboards. Python **3.10+**.

Machine learning pipeline for predicting ICU sepsis mortality using the MIMIC-IV clinical database.