Hi, I’m Andy Lee, a recent graduate of Renmin University of China, and I received my degree in computer science. I am part of the Turing Honors Class, a highly selective program designed to cultivate top-tier talent. The core of this program is not just the rigorous curriculum, but the environment. It places you in a high-pressure setting with extremely talented peers.
This intense competition conditioned me to learn complex technologies quickly and thrive under pressure, which was crucial for adapting to the fast-paced research culture at Berkeley. I was thrilled to be accepted into the Fall 2024 cohort of the Berkeley Global Access (BGA) Program!
Berkeley is the epicenter of open-source systems research. I chose BGA to trace the source of the technologies that are changing the industry. I wanted to understand the origins of global standards like Spark, Ray and vLLM. The program offered the perfect platform to do this. Even just sitting in the classroom, I could feel the pulse of innovation. The program gave me the chance to see firsthand how a research idea on a lab whiteboard evolves into a system used by millions.
Curated Course Selections
My strategy was vertical integration. I took Great Ideas of Computer Architecture (Machine Structures) COMPSCI 61C to build a rock-solid foundation, thoroughly connecting the logic from transistors and assembly language all the way up to the hardware-software interface.
Simultaneously, I challenged myself with the graduate seminar COMPSCI 294 on machine learning systems. This course took me into the unknown, where we moved beyond textbooks and into debating the latest papers every week, exploring problems that don't have standard answers yet.
My experience getting into the graduate seminar is a fond memory. On the first day, the room was so packed that many students, including myself, were sitting on the floor. The professor warned us that due to capacity, most undergraduates probably would not be able to enroll. However, I decided to have a thick skin, so I kept sitting on the floor and attending every lecture. That persistence paid off. I eventually got into the class, which became my gateway into the Sky Computing Lab.
These courses taught me that abstraction is power.
In that class, my final project explored Directed Acyclic Graph (DAG) scheduling for AI workloads. The real challenge was not just writing code, but designing an elegant interface to hide the underlying complexity. While the coursework code was a prototype, the architectural insights I developed there later evolved into the production-grade Job Groups feature in SkyPilot, which is an open-source framework I contributed to.
Seeing my classroom ideas translate into features used by thousands of developers taught me that making a system simple is often harder and more valuable than just making it powerful.
I challenged myself with the graduate seminar COMPSCI 294 on machine learning systems. This course took me into the unknown, where we moved beyond textbooks and into debating the latest papers every week, exploring problems that don't have standard answers yet.
Berkeley Mindset
In the very first lecture of the seminar, Professor Ion Stoica said that finding the right problem is the hard part. Because everyone at Berkeley is smart, the solution is often the easier part.
This made me realize that research is essentially betting. At Berkeley, we are trained to look three to five years ahead and predict future trends. Because it is a bet, you have to be willing to accept failure or temporary setbacks. This willingness to embrace risk and focus on long-term impact—rather than just solving immediate textbook problems—was a paradigm shift for me.
Soaking Up the Sun
Nature is my best debugger.
I absolutely loved the California sunshine. It called me out every day to recharge. The only downside is the winter rain, which makes hiking a bit muddy, but the good days far outweigh the bad.
My favorite spot is the Dipsea Trail. The contrast there is unbelievable. One moment you are hiking through open grasslands, the next you are in a lush, rainforest-like canyon, and finally you pop out to a view of the vast ocean.
I also enjoyed the challenge of Mission Peak. It is a steep and exposed mountain, but the sense of achievement at the summit resets my brain.
My top five things to do in Berkeley were:
- Spending late nights at Soda Hall collaborating with lab mates to meet a paper deadline.
- Watching the breathtaking sunset from the hill behind Soda Hall around 5 pm.
- Exploring Mission Peak for the pure physical challenge.
- Utilizing the tech resources as the density of brilliance is high and resources are everywhere, with constant high-quality tech talks.
- Eating at House of Three restaurant, which is right next to Soda Hall. The food is a great comfort after experiments, and the owners are incredibly nice people.
Berkeley is the epicenter of open-source systems research. I chose BGA to trace the source of the technologies that are changing the industry.
Coaching Leads to Ph.D. Program!
I also took the graduate school coaching course with Sally Clapper, and it was incredibly helpful. Sally was very responsible and helped me polish the very first version of my CV.
The program also invited professional speakers who taught us essential career skills for life in the U.S., such as the art of the coffee chat. These networking skills gave me the confidence to reach out to professors and Ph.D. students, laying the groundwork for my research journey. This support system was instrumental in my recent admission to the Ph.D. program at Princeton University!
When I look into my future, I hope to be a system builder. Whether in academia or industry, I want to build the next generation of AI infrastructure, helping to democratize AI computation so that it becomes as accessible as electricity.
My Advice for You
Don't build toys—aim for adoption.
Don't just write code for a grade. At Berkeley, adoption is the true metric of research impact. For example, my project, LEANN, started as an idea but grew into an open-source library with more than 10k GitHub stars.
So my advice is to embrace permission-less innovation. You do not need anyone's approval to contribute to a major open-source project or solve a real problem.
Break the hierarchy, reach out to people and build something real.
