Hi! This is Shu. I am a Ph.D. student in Computer Science at UC Berkeley Sky Computing Lab, advised by Ion Stoica. Before Berkeley, I received my B.S. in Computer Science and Applied Math from the University of Wisconsin – Madison, where I was fortunate to be advised by Aditya Akella and Shivaram Venkataraman.
My research lies at the intersection of AI and systems, with a focus on methods and infrastructure for self-improving agents. I build test-time scaffolds, post-training methods, and systems infrastructure that let agents solve long-horizon coding tasks, improve real-world software, and accelerate scientific and systems discovery. I recently work on SkyRL, a modular full-stack RL library for training LLM agents, and SkyDiscover, an adaptive framework for AI-driven discovery.
I’m always happy to connect with students interested in research or collaboration. If you’re interested, please send me an email at lshu [at] berkeley [dot] edu.
EvoX: Meta-Evolution for Automated Discovery
COLM 2026.
[Arxiv]
[code]
SkyDiscover: A Flexible, Adaptive Framework for AI-Driven Scientific and Algorithmic Discovery
CAIS 2026. ⭐ Industry Spotlight
[paper]
[blog]
[code]
LEANN: A Low-Storage Vector Index
MLSys 2026. ⭐ Best Paper Award
[Arxiv]
[code]
FrontierCS: Evolving Challenges for Evolving Intelligence
ICML 2026.
[Arxiv]
[code]
LLMs Can Easily Learn to Reason from Demonstrations – Structure, not content, is what matters!
Findings of EMNLP 2025.
[paper]
[code]
Optimizing LLM Queries in Relational Data Analytics Workloads
MLSys 2025.
[paper]
[code]
SkyStore: Cost-Optimized Object Storage Across Regions and Clouds
VLDB 2025. ⭐ Invited to VLDB Journal “Best of VLDB 2025”. Deployed at IBM.
[paper]
[code]
Throughput-Oriented MoE Inference on Memory-Constraint GPUs
ASPLOS 2025.
[paper]
[code]
JENGA: Effective Memory Management for Serving LLM with Heterogeneity
SOSP 2025.
[paper]
[code]
Cloudcast: High-Throughput, Cost-Aware Overlay Multicast in the Cloud
NSDI 2024
[paper]
[code]
Darwin: Flexible Learning-Based CDN Caching
SIGCOMM 2023
[paper]
[code]
Supporting Our AI Overlords: Redesigning Data Systems to be Agent-First
CIDR 2026
[paper]
SkyRL-SQL: Matching GPT-4o and o4-mini on Text2SQL with Multi-Turn RL
NeurIPS 2025, Multi-Turn Interactions in LLMs Workshop
[blog][code]
Barbarians at the Gate: How AI is Upending Systems Research
[Arxiv][code]
AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization
[Arxiv]
[code]
Pie: Pooling CPU Memory for LLM Inference [Arxiv]
Check out more recent preprints here!
I’m a big soccer fan: a lifelong FC Bayern supporter and a part-time Man United follower. I’ve been to many Bundesliga matches in person and collected several player autographs over the years. Mia San Mia!
I’m very into sports: running (5–10Ks and half marathon at Madison, Saarbrücken, and Berkeley), badminton, and both bouldering and rope climbing (indoors and recently outdoors).
I enjoy a mix of music, literature, and travel. I love playing the piano and listening to classical. I love reading: find a list of books I’ve read here. I enjoy traveling and photography. Check out some of the photos I took here!