Letian Ruan

 |  News  |  Experience  |  Publications  |  Projects  | 

I am a sophomore student at University of Michigan, majoring Data Science under Computer Science and Engineering Department. Besides, I am also dual majoring in Mechanical Engineering at Shanghai Jiao Tong University.

Currently, I'm focusing on the research of machine learning systems, large language model infrastructures and high performance computing. To be more specific, I'm currently working on improving LLM inferring by quantilizing model parameters for memory reduction, parallelizing computing tasks for acceleration and optimizing kernel operators for efficiency. Besides, I am fortunate to be advised by Shixuan Sun in Emerging Parallel Computing Center from SJTU-CSE.

CV  |  Email  |  GitHub  |  LinkedIn

profile photo

Mar '25  

 I am admitted to the University of Michigan, College of Engineering, Computer Science and Engineering Department.

Oct '24  

 I joined the Emerging Parallel Computing Center as a research assistant under supervision of Prof. Shixuan Sun
Undergraduate Student | University of Michigan, Ann Arbor
August '25 - Present

B.S.E in Data Science, enrolled in Computer Science and Engineering Department, College of Engineering.

Undergraduate Student | Shanghai Jiao Tong University
September '22 - Present

B.E in Mechanical Engineering, enrolled in UM-SJTU Joint Institute.

FaaSBoard: Efficient Graph Processing with a Disaggregated Architecture on Serverless Services

Under submission. [preprint]

We analyze the limitations of monolithic function architectures in graph processing, and motivate the design of a disaggregated serverless architecture. We introduce FaaSBoard, a graph processing system built on serverless cloud services to validate the effectiveness of the disaggregated architectural design. We conduct a comprehensive evaluation of FaaSBoard to demonstrate its superior performance and cost.

The largest unofficial course review platform of SJTU

Open-sourse project. [website]

The website service is based on the Next.js and Gin, integrated with Vite, Redis, PostgreSQL and LLM to provide more personalized and faster reponses. As of November 2024, more than 28,000 users have posted over 33,000 reviews for 4,500+ courses.
Complete row-store relational database system

Course project of 15-445. [code]

This database management system implemented a buffer pool to improve memory utilization and developed Extensible HashTable for indexing and used Crabbing Protocol to add read-write locks. Besides, it also completed basic SQL operators and optimizers on the volcano model and built a centralized transaction manager based on the MVCC protocol, supporting snapshot isolation.
Simple distributed file system similar to Google File System

Course project of SE3331. [code]

This system has constructed a basic single-machine inode-based filesystem and extended it into a robust distributed filesystem with file splitting, lock manager and crash recovery. Also, it has built a MapReduce framework and achieve the Raft consensus algorithm of leader election, heartbeat mechanism, log replication protocol and snapshot mechanism.

This template is a modification to Jon Barron's website.