Udaya Sai Chirumamilla

Designed and deployed enterprise-scale ML and Generative AI solutions for healthcare claims, member, and provider analytics. Developed claims risk scoring and ensemble models, and built fraud/waste/abuse detection using supervised and unsupervised techniques. Engineered RAG pipelines using Azure OpenAI/GPT-4, LangChain, Azure AI Search, embeddings, and hybrid retrieval to improve document intelligence and retrieval performance. Built healthcare NLP pipelines with Transformers and SpaCy for classification and information extraction, implemented SHAP/LIME explainability, and established LLM/RAG evaluation and guardrails for grounded, low-hallucination responses. Delivered production real-time inference services using FastAPI/Azure Functions/AKS, and set up MLOps with Azure ML, MLflow, and Azure DevOps for automated training, deployment, monitoring, drift detection, and alerting. Utilized streaming and large-scale processing with Event Hubs, Databricks, ADLS Gen2, Delta Lake, and Synapse

Built production-grade payment transaction risk and fraud models using Python, Scikit-learn, XGBoost, and AWS SageMaker. Developed anomaly detection systems (Isolation Forest, Autoencoders) and behavioral analytics to identify suspicious transaction patterns while reducing manual investigations. Engineered scalable feature processing pipelines with PySpark, AWS Glue, EMR, S3, and Redshift to reduce data processing time and create model-ready datasets. Applied NLP and Transformer-based techniques (SpaCy, Hugging Face, BERT, AWS Comprehend) for document/customer interaction classification and information extraction. Implemented customer and merchant segmentation with clustering and dimensionality reduction, and created scalable training/deployment workflows using SageMaker, Docker, Step Functions, Lambda, and CI/CD automation. Delivered real-time scoring services via REST APIs and added monitoring/explainability using SHAP/LIME with CloudWatch and Grafana to track drift, latency, and fai

Developed production machine learning solutions for payment transaction analytics, focusing on high-risk pattern identification using Python, Scikit-learn, XGBoost, and AWS SageMaker. Built fraud/anomaly detection models with Isolation Forest, XGBoost, and Autoencoders, and created transaction risk scoring using historical behavior, merchant attributes, and temporal features. Engineered scalable feature processing pipelines using PySpark, AWS Glue, EMR, S3, and Redshift to generate ML-ready features and reduce processing time. Applied NLP with SpaCy, NLTK, Transformers, and AWS Comprehend for classification and extraction from payment-related documents. Created BERT/Transformer embeddings for semantic similarity tasks and performed customer/merchant segmentation with clustering and dimensionality reduction. Implemented containerized training/inference and automated retraining/deployment pipelines with SageMaker, Docker, Step Functions, Lambda, and monitoring with CloudWatch/Grafana. Ad