Building the things I imagined as a kid.

Train temporal convolutional networks on OpenSim inverse-dynamics targets for a powered knee-and-ankle prosthesis, running inside the real-time control loop on its embedded Linux controller.

Quant developer/analyst on the $3.1M student-managed fund, building backtesting infrastructure and trading strategies.

Owned the agent layer of a data platform University of California campuses run analytics on: 21 MCP tools over a 29,980-node dependency graph, 45x less context per call, and agent reliability raised from 83% to 100%.

Feature-engineered truth and other modality directions in LLMs, steering with XGBoost to improve on linear-probe and contrastive-mean-difference baselines.

Trained a supervised CNN in PyTorch on multi-year satellite imagery (macro F1 0.758) and served it from a containerized Flask app on GCP, on a four-person team.

Built a sports prediction app on a four-person team: a multithreaded C++17 service sustaining 50K+ ops/sec at sub-millisecond latency, and a React Native client with 500+ monthly active users.

Built enterprise applications in Java, Spring Boot, and MySQL for a global Oracle Cloud rollout at an 800+ consultant firm.