Dharani Satwika Komaravolu

Built RAG pipelines using Azure OpenAI, LangChain, and FAISS for clinical information search, reducing p95 response latency by 180 ms via metadata filters and caching. Improved clinical NLP inference using prompt compression, dynamic batching, and quantization to increase throughput by 20% while maintaining response quality. Automated clinical document classification with Azure OpenAI and NLP models, achieving 94% F1-score and reducing manual review effort by 30% using confidence-based routing. Deployed Docker and Kubernetes services serving 1M+ monthly API requests, increasing deployment frequency by 25% through CI/CD checks and standardized release management. Created validation pipelines in Python/Pandas/SQL for 500K+ monthly clinical records, reducing downstream quality issues by 28%. Tracked 20+ model iterations with MLflow and Azure Machine Learning, cutting release preparation time by 35%.

Built PySpark and Apache Spark ETL pipelines for 2M+ monthly records, reducing transformation time by 20% using partition pruning, broadcast joins, and optimized DataFrames. Trained XGBoost classification models with scikit-learn and feature engineering, improving prediction results by 15% on production customer risk datasets. Automated Python preprocessing and validation for 600K+ customer records monthly, reducing analyst preparation work by 30% through reusable shared transformations. Deployed machine learning endpoints with FastAPI, Docker, and CI/CD across 10+ services, reducing release turnaround from hours to under one hour. Created training pipelines for 1M+ historical records, reducing feature generation time by 25% through reusable optimized transformations. Used MLflow to track experiments and model versions across three client environments, reducing model comparison and release effort by 30% with consistent metrics and artifacts.