Yuri Takigawa

Portfolio / Blog / News

Yuri Takigawa

About

I belong to the Department of Creative Informatics in the Graduate School of Information Science and Technology at the University of Tokyo, and I am a first-year master's student in the Taura Lab. My main interest is high-performance computing, especially parallel and distributed computing.

Outside research, I enjoy watching sports. I support Kashima Antlers in the J.League and Arsenal in the English Premier League, and I also follow Major League Baseball. I also like driving, traveling, and watching films.

Research

I am interested in abstractions and software systems, including programming language implementations, that help users write high-performance programs more easily on massively parallel hardware.

Currently, I collaborate with Cerebras on algorithm research and on optimization and analysis frameworks for their unique hardware architecture, which has shown strong performance in LLM inference.

Keywords: distributed memory, Spatial Dataflow Architecture, LLM Kernel, Critical Path Analysis, Programming Model, memory allocation, register allocation, synchronous/asynchronous communication

Other Areas of Interest

LLM model architectures and optimization of inference and training workloads

  • Many existing models and workloads are designed around the memory-hierarchy assumptions of GPUs.
  • By relaxing hardware assumptions, there may be room to discover models or workloads that are both more accurate and faster.

High-level language systems that hide low-level, hardware-specific programming

  • Writing and debugging acceptable low-level programs on new hardware often comes with very high cost.
  • If resource management can be hidden from users, they can focus on expressing algorithms rather than device details.

Load-balancing algorithms for large-scale computation across multiple devices and nodes

  • A coordinated design across scheduling and data placement, both static and dynamic, is important in these environments.

Publication

TBA

Works

I publish selected work from research and engineering projects here. Click a card to see a popup with more detail.

Click any card to open a modal view.

Skills

Programming Languages

  • C/C++/Python/TypeScript
  • Minimal C compiler experience
  • Knowledge of register allocation
  • Assembly-level coding
  • Experience constructing a SIMD-vectorized natural logarithm on Cerebras ISA while correctly handling de-normals, NaN, and Inf

HPC

  • Working knowledge of CUDA runtime features
  • Experience measuring GPU kernel performance with NVIDIA profilers such as Nsight Compute and Nsight Systems
  • Able to write simple kernels in Triton
  • Able to write OpenMP for both CPU and GPU offload
  • Knowledge of fast LLM inference workloads and distributed training methods

Web Service

  • Worked at OneStepS on a SaaS platform integrating meeting recording capture, minutes generation, and AI slide generation.
  • Handled a distributed system spanning React frontend, AWS Lambda + DynamoDB + S3 + SQS APIs, a meeting bot, a FastAPI audio processing service, and a Bedrock-based slide generation service.
  • Contributed to the asynchronous pipeline from meeting URL intake through recording capture, transcription jobs, speaker diarization, minutes generation, and AI slide generation.
  • Participated in design, implementation, and operations in a team of about five people from the early phase of development to six months after release.

Computer Architecture

  • Experience designing and implementing an in-order pipeline processor and FFT-specific hardware from RTL through synthesis and place-and-route using Verilog / SystemVerilog.
  • Used Vivado, Synopsys Design Compiler, and Cadence Innovus in that work.

Misc.

  • More than one year of experience in an international collaborative research project conducted in English
  • TOEFL 91/120 in March 2025
  • Applied Information Technology Engineer Examination in October 2024
  • Tokyo Football Association referee grade 3 (active for 5 yrs)

Biography

Education

  • Apr 2019 - Mar 2022: Kaisei High School
  • Apr 2022 - Mar 2026: The University of Tokyo, College of Arts and Sciences / Department of Information and Communication Engineering

Work Experience

Apr 2022 - Present: Tokyo Educational Research Institute / Tetsuryokukai

  • Physics instructor for upper secondary classes and teaching materials
  • Information studies instructor for upper secondary classes and teaching materials

Dec 2024 - Present: OneStepS, ML / Software Engineer Intern

Resume

You can download my resume as a PDF.

Download Resume PDF