• Designed an enterprise RAG platform combining semantic retrieval and LLM reasoning, accelerating fraud investigations by 40% for 50+ analysts.
• Architected scalable RAG pipelines using FAISS, Pinecone, and ChromaDB, enabling sub second search across large financial document repositories.
• Developed agentic AI workflows using LangChain and LangGraph to automate multi step reasoning for fraud policy analysis.
• Enhanced fraud narrative classification accuracy by 17-20% through BERT fine tuning with LoRA and zero-shot classification techniques.
• Reduced fraud false positives by 18% by building ensemble risk models using XGBoost, Random Forest, and LightGBM.
• Implemented near real time credit risk scoring pipelines integrating transaction and bureau data.
• Deployed anomaly detection models using Isolation Forest to strengthen AML monitoring and detect suspicious activity.
• Automated large scale ETL pipelines using Apache Spark, AWS Glue, and S3 to process multi terabyte financial datasets.
• Operationalized ML services using FastAPI microservices, Docker containers, MLflow, and CI/CD pipelines.
• Integrated model monitoring and explainability frameworks using SHAP and LIME to support regulatory compliance.