Nevin Liang
Engineer
I'm a graduate student in computer science. I've worked on firmware, software, and computer vision at Verkada, Tesla, and HockeyStack.
Projects
Exploring
Quantum information
Error correction, resource estimation, circuit compilation, quantum complexity.
Learning theory
PAC learning and VC dimension, Rademacher complexity, concentration in high dimensions, online convex optimization and bandits, minimax lower bounds.
Scientific computing
Spectral methods, Hamiltonian simulation, and many-body dynamics for quantum chemistry and materials.
Building
nvglow
NVIDIA GPU benchmark and telemetry TUI for Linux. Named per-workload measurements rather than one collapsed score.
Agent Sandbox
Replay-first sandbox host for enterprise-agent research. Records human workflows in real apps, replays them locally instead of the real backend, and compiles recordings into workflow graphs.
Bourne Agents
AI revenue coworker built by the HockeyStack team. Chat, automations, shared deal memory, and a public API and MCP over the context layer.
Shipped
creeptoe
Crypto paper-trading simulator. Live Binance order-book reconstruction, execution simulation with configurable latency and fees, ClickHouse backtesting.
Marketing Blueprints
Identifies the engagement mix behind closed-won deals by segment, then scores how close each account is to that ideal.
Odin + Golden Paths
AI analytics assistant for HockeyStack dashboards. Trend and anomaly analysis, interactive Q&A, text-to-report, and Golden Paths journey mining.
Background
HockeyStack · Founding Engineer
Built Odin, agents, ensemble predictive models, Markov-model attribution.
NECL Labs, UCLA · Researcher
EMG signal capture and processing from the forearm for fingerless typing, including work with transradial amputees.
Verkada · Software and Firmware Engineer
Alarm Console, person detection, control-loop firmware, embedded Android.
Tesla · Embedded Systems Intern
FreeRTOS development tools for Model S Plaid and Cybertruck infotainment debugging.
BigML Labs, UCLA · Researcher
Data-efficient deep learning with convex loss functions.