OpenGym Copilot
A reward-tuning copilot for MuJoCo and OpenAI Gym environments. Plug in your own backend APIs and have fun tuning! Currently a prototype.
Hey, I'm Aaron 👋
I build robots, train models, and make pixels move.
I'm an EECS student at UC Berkeley working at the intersection of robotics AI & LLMs and computer graphics. Currently a software engineer on autonomy behavior evaluation at Nuro.
Off the keyboard, you'll find me at the gym, playing video games, or out on a hike.
🦾 robotics
🧠 LLMs
🎮 graphics
01 Projects
Things I've built, broken, and rebuilt.
Quest 3 → UR7e Teleop
Led the creation of a real-time teleoperation and replay system for the UR7e robot arm, driven by a Quest 3 VR headset.
IsaacSim Somersault
Simulations in Isaac Sim teaching a Unitree humanoid robot to do somersaults.
Low-Latency AI Guardrails
A method for building AI safety guardrails that cut latency by 99% without sacrificing performance, tested on an open-source safety dataset.
Cellular Automata Game
A cellular-automata-based voxel game system (think Minecraft) built for Berkeley's computer graphics course.
NOME / JIPCAD
Built a VSCode extension for the NOME language, designed knot structures, and implemented shelling, merging, offsetting, and color-changing in this open-source 3D CAD tool.
Little tools
02 Experience
Where I've been spending my days.
-
2026
Software Engineer, Autonomy — Behavior Evaluation
Nuro
- Building data cleaning, preprocessing, and analysis pipelines.
- Productionizing model-based approaches for evaluating autonomous vehicle behavior.
- Working on Eval for World Models
-
2025 – 2026
Researcher
Berkeley Sky Computing Lab · SkyRL
- Built tooling to merge sharded FSDP model tensors in each RL training checkpoint into a single
.safetensorsfile. - Integrated SkyRL-train into rLLM as a supported backend trainer through a new API.
- Built tooling to merge sharded FSDP model tensors in each RL training checkpoint into a single
-
2024 – 2025
AI Scientist
Uniphore
- Developed AI safety methodologies, including novel guardrail models.
- Developed versatile topic classifiers.
-
2023 – 2024
Peer Leadership Consultant
UC Berkeley · Organizational Advising & Student Involvement Services
- Processed official registration of student orgs and their signatories.
- Staffed the front desk answering questions from student orgs.
- Helped design a new training website for signatories.
-
2023
Language Engineering Intern
Ivoyant
- Built and debugged parsers and language models to analyze JSON files of user data.
- Built a data-mapping app feature that maps source to target schemas from parsed user-input expressions.
03 Academics & Writing
Papers and course projects I'm proud of.
Publications
- Low-latency AI safety guardrails COLING 2025
Course projects
- Deep Neural Networks CS 182
- Digital Design & Integrated Circuits EECS 151
- Computer Vision CV
- Computer Graphics CS 184
- LLM Agents Paper
04 Now & Interests
What I'm into these days. TODO: fill me in
📚
Reading
Papers or books on your nightstand — e.g. something on robot learning, RL for LLMs, or a good sci-fi novel.
Read more →🧪
Currently exploring
Research directions you're curious about — e.g. VLA models, sim-to-real, world models, evals for AV behavior.
Read more →🏋️
Training
Current lifting split, a PR you're chasing, or a goal for this year.
Read more →🥾
Favorite trails
Bay Area hikes you'd recommend — Mission Peak? Mt. Tam? Your call.
Read more →🎮
Playing
Games in your rotation right now.
Read more →🍵
Tea order
Your go-to drink for those tea chats.
Read more →Let's chat over tea 🍵
Robotics, LLMs, graphics, or a good hiking trail — I'm always happy to talk.