Sunil Janghu

I’m a Senior Associate in Data Science (Fraud & Machine Learning) at National Payments Corporation of India, focused on production-grade fraud detection for UPI transaction streams. I engineer models by ensembling LSTM sequence networks with XGBoost gradient-boosted trees over rolling 30-day customer histories.

I build automation that makes teams faster and models stronger. I architect an automated hyperparameter optimization framework that benchmarks ROC-AUC, precision-recall, and lift, then auto-selects optimal parameters for downstream training—plus I delivered a LangGraph-powered agentic AI assistant that turns natural-language briefs into end-to-end ML workflows.

Across my projects, I’ve shipped hybrid deep-learning fraud detection that balances performance and user experience. In a two-stage LSTM + XGBoost pipeline, I flag fraudulent UPI payments under 0.1% positive-class imbalance, and it was deployed for production scoring millions of transactions daily with sub-second latency.

I also build GenAI systems that unlock self-serve operations. I designed multi-agent feature engineering automation with LangGraph and Qwen-32B to reduce analyst turnaround from days to under one hour, and I built an enterprise RAG platform using LangChain and ChromaDB to cut document lookup time from minutes to seconds.