Readmit-IQ
End-to-end ML system for hospital readmission risk. PySpark medallion architecture, scikit-learn modeling with SHAP explainability, FastAPI inference, and a React frontend with RAG over PubMed via Qdrant.
- PySpark
- scikit-learn
- SHAP
- FastAPI
- React
Role
Data engineering, modeling, explainability, application build
Timeframe
2025
The problem
30-day readmissions are costly and partly preventable, but care teams need risk scores they can actually interpret and act on — not a black box.
Approach
- 01
Built a medallion (bronze/silver/gold) pipeline in PySpark to clean and standardize encounter, diagnosis, and demographic data.
- 02
Engineered features across utilization history, comorbidities, and length of stay, with leakage checks at each step.
- 03
Trained and calibrated scikit-learn classifiers, then used SHAP to explain individual predictions.
- 04
Served the model through FastAPI and paired it with a React interface plus RAG over PubMed (Qdrant) for supporting literature.
Outcome
Risk scores that come with per-patient driver explanations instead of an opaque number.
A reproducible pipeline that re-runs from raw data through to serving with a single command.
Clinical context surfaced alongside the prediction, so reviewers can sanity-check the model's reasoning.
More case studies
SaaS Retention Warehouse
A full analytics warehouse for SaaS revenue and retention. Built a star schema in DuckDB, modeled cohort retention, MRR waterfall, NRR, and LTV, and served metrics through a FastAPI + React dashboard.
Surgical Complications Benchmarking
Multi-page Streamlit dashboard benchmarking real CMS hospital surgical outcomes. Includes a calibrated synthetic patient-level layer for risk adjustment and predictive modeling.
Infection Control Surveillance
Streamlit dashboard surveilling real CDC/CMS hospital-acquired infection data. Features funnel-plot control charts, FDR-corrected hypothesis testing, and a predictive model.