Vivek Reddy Pulakannti

Developed enterprise-scale AI/ML solutions for claims analytics, fraud detection, risk scoring, underwriting optimization, and loss prediction. Built claims risk prediction and fraud detection models using scikit-learn, XGBoost, LightGBM, Random Forest, Isolation Forest, and anomaly detection techniques, including handling imbalanced datasets with SMOTE and threshold optimization. Implemented large-scale feature engineering pipelines with PySpark/Databricks/SQL for 20M+ records and improved processing efficiency. Built NLP pipelines with spaCy, Hugging Face Transformers, and BERT for entity extraction, text classification, and semantic similarity. Designed and deployed a RAG solution using Azure OpenAI, GPT-4, LangChain, Azure AI Search, and vector embeddings to enable faster and grounded retrieval, and created generative document intelligence workflows for summarization and policy interpretation. Implemented MLOps with MLflow and Azure ML, including CI/CD, automated validation, monito

Built machine learning models for credit risk assessment, loan default/delinquency prediction, portfolio analytics, forecasting, and financial risk management. Developed PD and challenger models using Logistic Regression, Random Forest, XGBoost, and gradient boosting, with rigorous evaluation (ROC-AUC, KS, Gini, Precision/Recall/F1). Engineered borrower-level features from credit/loan/payment/transaction/customer data using Python, SQL, Pandas, and PySpark. Implemented credit scorecard development techniques such as WOE/IV, variable binning, and monotonic transformations, plus stability and validation checks (PSI, sensitivity analysis) for regulatory alignment. Created anomaly detection for suspicious transactions using Isolation Forest and statistical methods. Set up MLOps workflows with Azure Machine Learning and MLflow for experiment tracking, model versioning, CI/CD, monitoring, and deployment. Delivered Power BI dashboards for portfolio KPIs and model metrics.

Developed machine learning and NLP solutions for clinical analytics, patient risk stratification, readmission prediction, disease progression analysis, and healthcare operational optimization. Built predictive models using scikit-learn, TensorFlow, PyTorch, XGBoost, Random Forest, and Logistic Regression, with feature engineering from EHR data (diagnoses, procedures, labs, medications, encounter history). Implemented NLP pipelines using spaCy, NLTK, and Hugging Face Transformers for classification, NER, semantic similarity, and document clustering to extract medical concepts. Applied BERT-based models for clinical text understanding. Conducted survival analysis and hypothesis/cohort evaluations, and used appropriate performance and calibration metrics. Added explainability via SHAP and feature attribution for clinician trust. Scaled processing with Spark/PySpark/Databricks and supported reproducible model management with MLflow.