Built SCx, the fleet operations system for Azure's AI supercomputers. Power, repave, and thermal operations across tens of thousands of GPU nodes (GB300, H200, H100). LLM inference for Copilot: millions of prompts per hour, sub-ms latency, GDPR-compliant.
Experience
Refactored distributed Go microservices and Terraform on AWS powering phishing and threat detection, cutting cloud spend ~75% (~$3M/yr).
Real-time video-understanding models powering Fitness Ally, 20BN's AI fitness coach. Company advised by Yoshua Bengio, acquired by Qualcomm.
Seven-month internship optimizing delivery speed promises across Amazon's EU fulfillment network. Modeled product placement and routing across fulfillment centers, distribution nodes, and last-mile handoffs to minimize delivery time, impacting the 1-day and 2-day delivery promises.
Built real-time ETF data visualization tools using D3.js and Python, supporting the launch of new financial products.
Selected for Huawei's Seeds for the Future program. Studied telecom infrastructure in Shenzhen and Beijing.
Replaced Excel workflows with scikit-learn and Keras models for APR price optimization on SMB transactional data.