Sashank Punyamurthy

Designed and deployed a machine learning-driven clinical risk prediction platform analyzing longitudinal EHR, claims and clinical notes via FHIR APIs and HL7 pipelines, enabling early identification of high-risk patients across chronic disease cohorts. Built scalable feature engineering and data ingestion pipelines with Apache Spark, Kafka and Databricks, processing patient records and clinical events, improving predictive model training throughput by 55%. Developed NLP pipelines for clinical text analysis using Spark NLP, BioBERT and ClinicalBERT to extract comorbidities, medication patterns and adverse events from physician notes, improving risk stratification model accuracy by 18%. Implemented medical imaging AI workflows using MONAI and PyTorch to support radiology-assisted diagnostics through automated anomaly detection across DICOM imaging streams from PACS systems, reducing manual review workload for radiology teams. Integrated predictive models into clinical applications throug

AI/ML Engineer building a Generative AI-powered healthcare/insurance assistant using Retrieval-Augmented Generation (RAG) and large language models. Implemented end-to-end document ingestion (PDF parsing, cleaning, chunking, embeddings), FAISS-based vector similarity search, and LangChain tool-based workflows. Developed a secure FastAPI inference service, containerized with Docker, and deployed on Kubernetes (basic) and AWS (S3, EC2/SageMaker). Applied prompt engineering and guardrails for policy-compliant responses with citation-style outputs and metadata-based filtering to improve recall. Tracked experiments with MLflow for reproducibility and managed model iterations.

Developed an end-to-end Fraud Detection and Credit Risk Scoring system for fintech transaction datasets. Conducted advanced feature engineering (velocity signals, device/IP risk, user behavior, merchant risk) and addressed class imbalance via SMOTE, weighted losses, and threshold tuning. Built models with XGBoost and Scikit-learn, and designed batch/near real-time scoring pipelines. Implemented a scalable FastAPI inference service, and managed model lifecycle with MLflow. Automated training and deployment workflows using Apache Airflow. Created analytics dashboards in Tableau/Power BI to monitor fraud trends, drift, and risk metrics. Containerized components with Docker for consistent deployments.