Researcher at University of Pennsylvania

Automating robot reward design with LLMs

My method, FORGE, beat the prior best on humanoid locomotion and standup at 10× less compute per individual.

Talk to me about
Cadmus

Starting a marketplace where AI agents shop

Agent payment rails got built, but nobody built the trust layer. We're starting with evals and routing for paid tools.

Talk to me about
Researcher at University of Pennsylvania·2026

Showed no learned scheduler can beat FCFS

A four-symmetry analysis tied our results to queueing theory and found which symmetry-breaks would change that.

Talk to me about
工学学士学位 in 化学工程 at Zhejiang University·2021 – 2025

Won 2nd prize in national chem-eng design

I studied chemical engineering, and in 2023 took third in Zhejiang's provincial physics innovation competition.

Talk to me about
Linqi Zhe

Linqi Zhe⁠

University of Pennsylvania

Researcher at University of Pennsylvania

About

I research automated reward design for reinforcement learning and robotics, and I run Cadmus, a marketplace for agents.

Experiences

  1. Researcher at University of PennsylvaniaAutomating robot reward design with LLMs

    My method, FORGE, beat the prior best on humanoid locomotion and standup at 10× less compute per individual.

  2. CadmusStarting a marketplace where AI agents shop

    Agent payment rails got built, but nobody built the trust layer. We're starting with evals and routing for paid tools.

  3. Researcher at University of PennsylvaniaShowed no learned scheduler can beat FCFS

    A four-symmetry analysis tied our results to queueing theory and found which symmetry-breaks would change that.

  4. 工学学士学位 in 化学工程 at Zhejiang UniversityWon 2nd prize in national chem-eng design

    I studied chemical engineering, and in 2023 took third in Zhejiang's provincial physics innovation competition.

Follow Linqi for more...

LinkedInX

© 2026 Delphi · Terms · Privacy

Linqi Zhe

Linqi Zhe⁠

University of Pennsylvania

Researcher at University of Pennsylvania

About

I research automated reward design for reinforcement learning and robotics, and I run Cadmus, a marketplace for agents.

Experiences

  1. Researcher at University of PennsylvaniaAutomating robot reward design with LLMs

    My method, FORGE, beat the prior best on humanoid locomotion and standup at 10× less compute per individual.

  2. CadmusStarting a marketplace where AI agents shop

    Agent payment rails got built, but nobody built the trust layer. We're starting with evals and routing for paid tools.

  3. Researcher at University of PennsylvaniaShowed no learned scheduler can beat FCFS

    A four-symmetry analysis tied our results to queueing theory and found which symmetry-breaks would change that.

  4. 工学学士学位 in 化学工程 at Zhejiang UniversityWon 2nd prize in national chem-eng design

    I studied chemical engineering, and in 2023 took third in Zhejiang's provincial physics innovation competition.

Follow Linqi for more...

LinkedInX

© 2026 Delphi · Terms · Privacy

By using this service, you agree to the Terms of Service and Privacy Policy.