02 — SELECTED WORK
Built a GPU-accelerated Monte Carlo engine to model stock price distributions from historical market data.
Uses geometric Brownian motion and independent cuRAND sampling with 10 years of historical trading data. The project benchmark compares 1,748 ms on a single CPU thread with 35 ms on an NVIDIA T4 for 1 million simulations over 30 days.
Built a dependency-free PCAP analyzer that decodes network protocols, recovers DNS hostnames, and groups packets into bidirectional flows.
Decodes Ethernet, IPv4, TCP, and UDP directly from packet bytes. Includes defensive bounds checks, capture byte-order detection, and rejection of unsupported formats. The resume benchmark processes 5,000 live packets (approximately 4.3 MB) into 13 flows in 15 ms.
Implemented graph algorithms to identify connected neighborhoods, find routes, and minimize utility-grid wiring in a city road network.
Uses union-find with path compression and union-by-rank for connectivity, modified Dijkstra routing to avoid selected intersection types, and Kruskal’s algorithm for a minimum-wiring utility grid.