About

I'm Chuyue (Steven) Wang, a computer science undergraduate at the University of Pennsylvania, Class of 2029. I work on LLM inference and agent systems: how to serve large models faster and at lower cost, and how to build agents that can be trusted with real work.

At Penn's Distributed Systems Lab I work on sparse attention that cuts KV-cache reads without losing accuracy, and on a memory layer that lets an agent reuse what it has already worked out. What keeps me in this work is that the hard part is rarely the idea. It is the latency budget, the stale state, and the instruction that only half specifies the job, and getting those right is what turns a demo into something people can rely on.

Most of what I build runs on SGLang and vLLM, and I contribute to both upstream. I believe the infrastructure behind these models should be open and well understood, and that a problem worth solving is worth solving at the source, so the next person never meets it. That is the work I want to keep doing.

Experience

Internships

Aligned Agent Inc. (Kosa)

Aug 2026 – Present

Software Engineering Intern · New York, NY

Shipped MiniKosa, a macOS copilot with on-device transcription, and built the commit layer behind Kosa's deal updates.

Hongcheng Big Data Institute × Nanjing University

May 2026 – Aug 2026

Research Engineering Intern · Nanjing, China

Co-authored Traj2Mem, a procedural memory that lets an agent reuse its past trajectories, and owned its retrieval and evaluation.

Paperboy Inc.

Mar 2026 – Apr 2026

Agent Engineering Intern · San Francisco, CA

Made the Engine agents' tests deterministic with a record-and-replay harness.

Franklink Inc.

Oct 2025 – Feb 2026

Founding Engineer · Philadelphia, PA

Built the backend of an AI networking agent inside iMessage: LangGraph orchestration, hybrid matching, and durable queues.

Huatai Securities

Mar 2024 – Jun 2024

Derivatives Valuation Intern · Nanjing, China

Implemented Gaussian HJM models for swaption pricing and rewrote the desk's daily exotic-equity valuation, cutting a run from two days to four hours.

Research

Research Engineer · Philadelphia, PA · advised by Prof. Bingxin Zhao

Designed a post-cutoff benchmark for LLM paper review that tests whether a model can judge a paper it has never seen.