Chaitanya Prasad Reddy Narala

Built real-time risk scoring and fraud-detection pipelines using Python, XGBoost, Kafka, and AWS SageMaker over 10M+ annual transactions, achieving sub-second scoring latency and reducing fraud losses by 45%. Developed an agentic LLM-based KYC/AML copilot using GPT-4o/Claude with LangChain-style multi-step retrieval to automate 70% of compliance workflows and reduce onboarding time by 60%. Engineered RAG pipelines with retrieval evaluation and guardrails to reduce hallucinations and improve investigator case-resolution speed by 50%. Implemented explainability and governance using SHAP/LIME, reducing false positives by 30% and producing audit-ready evidence for AML and SOC2. Productionized ML/LLM services with Docker, Kubernetes, MLflow, model registry, and CI/CD; also created synthetic fraud data to improve rare-event detection by 35%. Established LLM evaluation and guardrail testing (benchmarks and red-teaming) to validate quality and safety before release, improving GenAI reliability

Architected real-time fraud-detection pipelines using Kafka, Spark Streaming, and UPI/NPCI payment APIs at under 200 ms latency across 5M+ transactions, reducing fraud losses by 32%. Improved anomaly detection precision by 28% using XGBoost and feature engineering, lowering false positives in payment screening. Built fraud-scoring microservices with Spring Boot and Redis via REST APIs with 99.9% availability and sub-second risk scoring, and designed adaptive risk-rule engines to cut fraud leakage by 30% while supporting regulatory audits. Created AI-driven credit-scoring models using alternative data and bureau integrations (CIBIL/Experian), increasing loan approvals by 22% while lowering default exposure. Optimized ETL orchestration with Airflow and AWS S3, reducing loan-processing turnaround by 40% by accelerating KYC and verification, and delivered Tableau credit-risk dashboards on SQL pipelines to reduce delinquency by 18%. Implemented monitoring and data-quality checks and partner