Harshini Reddy

• Engineered real-time ML pipelines using Python, TensorFlow, Spark, Kafka, and feature engineering workflows to process 180,000+ daily transactions and improve risk-scoring throughput for customer-facing decision systems. • Designed personalization and recommendation experimentation frameworks across web and mobile, deploying 25+ A/B tests and increasing recommendation CTR by 12% and engagement by 8% through ranking, segmentation, and model evaluation. • Developed graph-based fraud detection models with PyTorch, Autoencoders, Graph Neural Networks, entity relationships, and Neo4j-style graph modeling concepts to identify 1,600+ anomalous events per month and reduce false alerts through feature-store optimization. • Built Customer 360 and context-aware analytics patterns by modeling accounts, customers, products, interactions, and risk events for downstream recommendations, knowledge graph use cases, and stakeholder reporting. • Operationalized the ML lifecycle with MLflow, Kubernetes, Airflow, CI/CD, model versioning, monitoring, drift detection, and data quality checks, supporting 30+ active production models with stronger reliability and governance. • Applied SHAP, LIME, fairness checks, and model-risk controls to explain model performance, improve Precision/Recall tradeoffs, and align AI/ML systems with regulated financial services standards. • Architected LangChain and retrieval pipelines for compliance intelligence, enabling auditors to retrieve insights in under 2 seconds across millions of records while supporting AI-assisted decision workflows.

● Developed credit-risk and default prediction models using XGBoost, LightGBM, and TensorFlow, powering underwriting for 20,000+ SME and retail applications per month and improving approval accuracy by 7%. ● Architected monitoring dashboards integrating ML predictions, drift metrics, and risk exposure across 16 business units using Tableau and Snowflake. ● Engineered real-time fraud detection pipelines with PyTorch, GNNs, Kafka, and Spark Streaming, identifying 1,500+ anomalous transactions/month while reducing false positives by 30%. ● Built NLP document intelligence workflows using Transformers, spaCy, and T5 to process 70,000+ financial cases monthly, improving routing accuracy. ● Optimized feature engineering pipelines with PySpark and Delta Lake, reducing model-preparation time from 12 hours to 4 hours across 2TB+ datasets. ● Developed time-series forecasting models (LSTM, Prophet, ARIMA) to predict volatility and liquidity across 150+ financial instruments. ● Productionized ML workflows using Docker, Kubernetes, MLflow, Airflow, and SageMaker, enabling CI/CD for 25+ models with automated retraining and monitoring. ● Designed AML/KYC risk engines combining ML and rule-based scoring, screening 300,000+ profiles and escalating 2,500+ high-risk entities. ● Applied explainability, bias detection, and drift monitoring to ensure compliance with SOX, Basel III, and OCC standards. ● Piloted generative AI retrieval systems using LangChain and embeddings, enabling compliance teams to query regulatory documents in under 2 seconds.