At Grab, I lead the MIWI Fault Attribution Engine for GrabFood, replacing a rule-based refund system with an AI-powered multimodal solution for post-delivery complaints. I built a GEPA agentic pipeline and a Pareto-based prompt selection framework across multimodal datasets.
At Visa Inc., I architected the ML pipeline for the Real-Time Payments Fraud System, achieving sub-100ms inference latency on millions of daily transactions. I also built an Agentic RAG intelligence layer that lifted new-client documentation coverage by 85%.
At Optum (UnitedHealth Group), I developed medical-claims risk models that increased detection accuracy by 15% and generated $2M+ in annual cost savings. I also fine-tuned Flan-T5 and Big Bird for Call Audit, increasing audit throughput fivefold while reducing manual review time by 80%.
