Yogitha Donepudi

Architected and delivered end-to-end AI/ML software solutions on AWS. Built scalable training and inference pipelines with SageMaker Pipelines and Endpoints, integrating Feature Store and Model Monitor. Developed prompt engineering frameworks and evaluation harnesses for generative AI, improving response quality across customer service, KYC, and compliance assistant use cases. Implemented real-time computer vision workflows with Amazon Rekognition for identity verification and fraud/document analysis, and built NLP pipelines with Amazon Comprehend and Hugging Face Transformers for sentiment, entity extraction, and PII redaction. Engineered ETL workflows using AWS Glue, EMR, PySpark, Lambda, and Step Functions, and containerized services deployed on EKS/Kubernetes with canary/rolling patterns and strong observability.

Built data pre-processing and feature engineering pipelines for large-scale clinical and claims datasets using Pandas, NumPy, PySpark, and SQL. Developed classification and regression models for patient risk scoring, disease prediction, and utilization forecasting. Trained deep learning models with TensorFlow/Keras and built NLP pipelines using BERT and spaCy for extracting clinical entities and information from unstructured text. Added model interpretability using SHAP and LIME and built validation frameworks with cross-validation and standard classification metrics. Delivered ML APIs using FastAPI/Flask, containerized applications with Docker, tracked experiments and models with MLflow, and supported lifecycle management for production deployments.

Designed feature engineering pipelines from transaction history, customer behavior, merchant patterns, geographic and temporal attributes to improve fraud detection and risk scoring. Built supervised and unsupervised models for fraud classification, anomaly detection, and customer segmentation, including Isolation Forest and clustering/anomaly approaches. Developed real-time ML scoring services using Python, Kafka, and REST APIs for low-latency transaction monitoring. Built scalable data processing workflows with Spark/PySpark and SQL, implemented NLP-based analysis over investigation notes and regulatory documentation, and established model monitoring for data/feature drift and production anomalies. Used MLflow for experiment tracking and model registry, and communicated results using dashboards and reports for engineering and compliance stakeholders.