RAJASEKHAR POLEPALLI

Built claim-note datasets using PySpark/Apache Spark and Azure Databricks with ADLS Gen2, applying preprocessing and feature engineering for claims-prioritization training. Trained prioritization models with scikit-learn and XGBoost using cross-validation, hyperparameter tuning, and explainability to reach 92% macro F1, enabling earlier associate review for higher-priority cases. Implemented MLflow experiment tracking and Model Registry workflows for reproducible model comparisons. Containerized real-time FastAPI inference services using Docker, Kubernetes, and Azure Machine Learning, improving latency by 21%. Added automated endpoint validation via Azure DevOps, CI/CD, and pytest to reduce deployment regressions by 16%. Worked with stakeholders on model monitoring and data/model drift analysis to drive targeted updates.

Developed a customer support chatbot using Azure OpenAI, LangChain, RAG, and prompt engineering, improving response relevance to 91% by grounding answers in approved FAQs and troubleshooting content. Organized 10K+ knowledge passages with vector embeddings and semantic search for retrieval. Fine-tuned BERT models for intent classification with PyTorch and Hugging Face Transformers, reducing incorrect support routing by 17%. Improved multi-turn behavior with context engineering and semantic search. Integrated retrieval and LLM responses through REST APIs. Applied LLM evaluation (groundedness), hallucination detection, PII protection, and AI guardrails to reduce unsupported or unsafe answers by 22%. Collaborated with ML, software, and QA teams via Git workflows to drive iteration based on feedback.