Mason Naung

A streaming fraud-detection pipeline built to mirror production payment-fraud scale and latency.

• -Architected an end-to-end streaming fraud detection pipeline processing 5,000+ TPS at sub-50ms P99 scoring latency, with XGBoost served from a Feast-backed online feature store using Redis sliding-window velocity counters.
• -Designed a graph-feature pipeline over a card-device-IP transaction graph using Louvain community detection, improving fraud recall at 1% FPR by 11 percentage points over a tabular-only baseline on the IEEE-CIS benchmark.
• -Shipped a champion-challenger shadow-deployment framework with Prometheus observability, Evidently-based drift detection, and cost-weighted threshold optimization for safe, iterative model rollout.

A visa-policy lookup platform for Refugee Travel Document holders, with an LLM-powered reporting pipeline.

• -Built a platform for US Refugee Travel Document (I-571) holders to check country-specific visa entry policies, serving 1,000+ monthly users via a 3D geospatial "Travel Freedom" map (D3.js, react-simple-maps) backed by a Neon serverless PostgreSQL database.
• -Applied GenAI/LLM techniques: traveler reports submitted via Telegram Bot are processed by Llama 3.3 (Groq) for structured policy extraction, then held in a human-in-the-loop admin approval flow before committing to Postgres.