Back to archive
MLOps
2025
ML engineer, team of 5

Healthcare Readmission Risk MLOps Pipeline

MLOps-focused healthcare pipeline that turns a notebook model into a reproducible training, evaluation, and inference system.

Converted a Jupyter notebook diabetes readmission model into a production-style, reproducible MLOps pipeline with experiment tracking, config-driven runs, and a Dockerised FastAPI /predict service.

Key Outcomes

9-stage reproducible pipeline
Minority-class recall improved from 17.5% to 57.6%
$15.79M projected savings from intervention simulation

Context

Problem and Context

Notebook-based healthcare models are difficult to reproduce, operationalize, and evaluate reliably, especially when severe class imbalance makes minority-risk detection the real business objective.

Approach

Approach and Architecture

A config-driven 9-stage MLOps pipeline that transformed a diabetes readmission notebook into a reproducible training, evaluation, and inference system with MLflow, W&B, Hydra, GitHub Actions, Docker, and a FastAPI /predict service.

Implementation

Implementation Details

Ingestion -> preprocessing -> feature engineering -> modelling -> evaluation -> inference across a 9-stage pipeline. Hydra orchestrates repeatable runs, MLflow and W&B track experiments and artifacts, GitHub Actions automates checks, and Dockerized FastAPI exposes synchronous inference patterns for real-time and batch-ready use.

Python
scikit-learn
MLflow
Hydra
W&B
Docker
FastAPI
GitHub Actions
BorderlineSMOTE

Results

Results and Tradeoffs

This project is presented as a concise technical overview rather than a full-length narrative case study.

9-stage reproducible pipeline
Minority-class recall improved from 17.5% to 57.6%
$15.79M projected savings from intervention simulation

Explore More

Related Projects

Browse adjacent work from the same archive group or jump back to the project archive.

Back to archive