I’m a systems generalist interested in systems for AI.
I work at SpaceXAI (formerly xAI), where I’m part of Inference and Grok Imagine. I also led Applied Inference, responsible for large-scale inference infrastructure. My work spans:
- Training: Grok Imagine upsampler — SFT, RL, and data
- Training for inference efficiency: Grok speculative decoding
- Inference: SGLang, scheduling, constrained decoding, load balancing, auto-scaling, and various products (API, search models, Grokipedia)
Before xAI, I was a researcher at SystemsResearch@Google, working on language model sparsity, long-context serving, and memory tiering — contributions include Gemma 3n and Google’s data center memory tiering systems. I was fortunate to work with David Culler, Hank Levy, Kim Keeton, and Arvind Krishnamurthy.
I hold a Ph.D. in Computer Sciences from the University of Wisconsin-Madison (2022, advised by Andrea and Remzi Arpaci-Dusseau) and a B.S. from USTC (2016). I’m also an open source contributor to SGLang, xgrammar, and LLVM.
Experience
SpaceXAI (formerly xAI)
SystemsResearch@Google
Microsoft Gray System Lab
VMware
Chinese University of Hong Kong
Publications
- MDK: Rethinking the Data Center Memory Reclamation Problem OSDI 202617th USENIX Symposium on Operating Systems Design and Implementation
- OBASE: Object-Based Address-Space Engineering to Improve Memory Tiering OSDI 202617th USENIX Symposium on Operating Systems Design and Implementation
- Spark Transformer: Reactivating Sparsity in FFN and Attention NeurIPS 2025Conference on Neural Information Processing Systems
- Getting the MOST out of your Storage Hierarchy with Mirror-Optimized Storage Tiering FAST 202624th USENIX Conference on File and Storage Technologies
- PageFlex: Flexible and Efficient User-space Delegation of Linux Paging Policies with eBPF ATC 20252025 USENIX Annual Technical Conference
- FineMem: Breaking the Allocation Overhead vs. Memory Waste Dilemma in Fine-Grained Disaggregated Memory Management OSDI 202519th USENIX Symposium on Operating Systems Design and Implementation
- SLAP: Segmented Reuse-Time-Label Based Admission Policy for Content Delivery Network Caching TACO 2024ACM Transactions on Architecture and Code Optimization
- Arya: Arbitrary Graph Pattern Mining with Decomposition-based Sampling NSDI 202320th USENIX Symposium on Networked Systems Design and Implementation
- SLAP: An Adaptive, Learned Admission Policy for Content Delivery Network Caching IPDPS 202337th International Parallel and Distributed Processing Symposium
- WiscSort: External Sorting for Byte Addressable Storage VLDB 202349th International Conference on Very Large Data Bases
- NyxCache: Flexible and Efficient Multi-tenant Persistent-Memory Caching FAST 202220th USENIX Conference on File and Storage Technologies
- Cornus: Atomic Commit for Cloud DBMS with Storage Disaggregation VLDB 202248th International Conference on Very Large Data Bases
- The Storage Hierarchy is Not a Hierarchy: Optimizing Caching on Modern Storage Devices with Orthus FAST 202119th USENIX Conference on File and Storage Technologies
- The Storage Hierarchy is Not a Hierarchy: Optimizing Caching on Modern Storage Devices with Orthus NVMW 202112th Non-Volatile Memories Workshop
- Releasing Locks As Early As You Can: Reducing Contention of Hotspots by Violating Two-Phase Locking SIGMOD 2021ACM SIGMOD International Conference on Management of Data
- Read as Needed: Building WiSER, a Flash-Optimized Search Engine FAST 202018th USENIX Conference on File and Storage Technologies
- Towards an Unwritten Contract of Intel Optane SSD HotStorage 201911th USENIX Workshop on Hot Topics in Storage and File Systems
- Exploiting Intel Optane SSD for Microsoft SQL Server DaMoN, SIGMOD 2019ACM SIGMOD International Conference on Management of Data
Service
Program committee: OSDI’26, MLSys’26, DIMES’25, FAST’25, ATC’25, ATC’24, HotStorage’24.
Also: USENIX HotStorage’24 Local Chair, SOSP’24 Proceedings Chair, SOSP’23 Student Research Committee & Poster Committee, ACM TOS’21–26 reviewer, VLDBJ’23 reviewer.