Poojitha Vajja

About

Built and deployed a real-time hospital readmission risk prediction system at NYU Langone Health, combining structured EHR data with BERT-based NLP on clinical notes and serving predictions to clinicians via Azure ML and FHIR APIs. Emphasizes production reliability and clinical trust through SHAP-based explainability and robust healthcare data preprocessing, and reports a 22% reduction in 30-day readmissions.

Experience

Education

Key Strengths

• Deployed real-time hospital readmission risk prediction system into clinical workflow via Azure ML + FHIR APIs
• Combined structured EHR data with BERT-based NLP on clinical notes for improved risk prediction
• Delivered measurable impact: reduced 30-day readmissions by 22%
• Strong handling of noisy healthcare data (code standardization, clinically meaningful imputations, low-signal feature filtering)
• Production-grade explainability using SHAP to provide clinician-auditable risk drivers at prediction time
• End-to-end ML pipeline orchestration with Airflow (scheduling, dependencies, retries) for scalable production workflows
• Reliability-focused evaluation and operations: success metrics, edge-case/failure testing, monitoring for drift/latency/errors, retraining loops
• Effective collaboration with clinicians; translates clinical goals into measurable ML outputs and iterates based on stakeholder feedback