Finn Sheng

Robotics · LLMs · AI Infra · Computer Engineering at NC State University

I work on robotics, large language models and the infrastructure that runs them: systems where a model can sense, reason about and act on the physical world. I build them end to end, from the serving stack down to the hardware. This site collects my publications, project write-ups and notes.

Finn Sheng

Experience

Jun 2026 – Aug 2026

AI Engineer · Espressif Systems

Lead the refactor on ESP-Claw, Espressif's official open-source AI agent framework: layered memory with automatic context compaction, a semantic finite-state machine in the agent loop, and a migration of the agent core from C to async Rust that sharply reduced memory use.

ESP-Claw · Rust · agent systems

Dec 2025 – Present

GPU Power Research Assistant · NC State University

Datacenter power forecasting that splits load into a time-series-predicted base and statistically estimated periodic components, and asynchronous checkpoint scheduling across multiple GPUs.

GPU power · LLM training

May 2025 – Jul 2025

Research Intern · Shanghai Jiao Tong University

Benchmarked LSTM and time-series foundation models (TimesFM) for multivariate forecasting under latency and memory limits, deployed and profiled on a Jetson Orin Nano.

Edge AI · forecasting

Projects

Lumen

Desk-first hardware presence on ESP32-C3.

Vision-UI

UI framework for small embedded displays.

VibeBox

Ultrafast agent sandbox for LLM agents.

Recent posts

Archive →

[Project] Lumen

A build log of Lumen, a desk-first hardware presence tying system state to motion, light, and sound through an ESP32-C3 firmware framework.

Project · ece, software

[Project] Vision-UI

A brief write-up on the architecture and decisions behind Vision UI.

Project