LLM Inference Systems
Studying scheduling, KV-cache behavior, SLO attainment, queue latency, and serving performance for interactive LLM workloads.
Building efficient AI systems for practical, trustworthy intelligence.
I am Zichao Lin, a Computer Engineering student at Zhejiang University and the University of Illinois Urbana-Champaign from 2022 to 2026. From 2026 to 2028, I will continue as a Computer Science student at UC San Diego, focusing on LLM inference systems, NLP applications, computer vision, and systems engineering.
- Focus
- AI Systems, LLM Inference, NLP
- Publication
- MLSys 2026
- Languages
- Mandarin, English
Research Direction
From model capability to deployable systems
NLP Applications
Applying LLMs to extract economic events, temporal logic, profitability signals, and data discrepancies from government reports.
Computer Vision
Exploring automated sports analytics with segmentation and action recognition, using SAM and Transformer-based model comparisons.
Selected Publication
SuperInfer at MLSys 2026
2026
SuperInfer: SLO-Aware Rotary Scheduling and Memory Management for LLM Inference on Superchips
J. Yu, M. Hu, Z. Lin, M. Zhang. Proceedings of the 9th Conference on Machine Learning and Systems (MLSys).
Contributions included multi-turn dialogue support, SLO monitoring, reproducible baseline setup, and visual analysis for KV-cache usage, queue latency, VLT growth, and SLO attainment.
Project Portfolio
Systems, AI, and teaching projects
Research Assistant | UIUC
Superchip LLM Inference Optimization
Extended vLLM and SuperInfer for multi-turn conversational workloads, monitored TTFT/TBT SLOs, and compared open-source LLM serving baselines under varied datasets.
- Diagnosed request-pressure effects on TBT under controlled workload changes.
- Produced research visuals for dynamic system metrics and performance evidence.
Research Team Member | ZJUI
Computer Vision Volleyball Analysis
Surveyed datasets and tools for automated match analysis, selected SAM for instance segmentation, and evaluated Transformer-based action recognition against ResNet.
- Documented model-dataset trade-offs and integration risks.
- Built foundations for performance evaluation and highlight generation.
Independent Researcher
RISC-V Operating System Kernel
Implemented a RISC-V file system, block device driver, virtual memory paging, system calls, fork, locking, and preemptive scheduling.
- Focused on process isolation, multitasking stability, and system responsiveness.
Independent Researcher
Dynamic Walk-Pairing and Path Planning
Designed a map-based route planning model with enhanced Dijkstra search, pyramid-tree candidate selection, and C memory management for large-scale requests.
- Generated matching results, optimal paths, and visualized map outputs.
Team Leader
Teaching TCP with Stop-Motion Video
Led a short-form teaching project explaining TCP handshake, cumulative ACKs, retransmission timeout, and congestion control through tangible visual metaphors.
- Built storyboards, motion graphics, subtitles, and hands-on communication props.
Experience & Leadership
Engineering discipline with community initiative
FinAI Intern, International Digital Economy Academy
Built LLM-based extraction modules for economic reports, integrated reranking and embeddings, and improved temporal information extraction through iterative prompting.
President, Zhejiang University Ultimate Frisbee Club
Founded and grew a campus ultimate community, organized weekly training and the Freshman Cup tournament, and led the team to a Zhejiang Division championship.
B.Eng. Electronic and Computer Engineering
Studying at Zhejiang University ZJU-UIUC Institute with an exchange experience at UIUC, taking coursework in algorithms, systems, machine learning, networks, and math.
Toolbox
What I work with
C
C++
Python
Linux
Git
GitHub
GDB
LaTeX
Google Cloud Platform
Machine Learning
Communication Networks
Operating Systems
Contact