Michio Sun
Hi! I'm a machine learning engineer based in Tokyo. I am generally interested in research and applications of AI for robotics, deep reinforcement learning, and ML optimizations.
Previously CS @ Tsinghua University in Beijing.
I am passionate about pushing the frontier of robotics. In my spare time, I enjoy running and making coffee.
Work Experience
EII Inc.
Sep 2025 - PresentTokyo, Japan
Machine Learning Engineer
- Led the company's electronics recycling automation initiative. Trained VLAs with Imitation Learning and value functions for RL in JAX, in collaboration with Waseda University and various municipal governments.
- Built end-to-end data collection and training pipelines spanning Franka, ALOHA, and UMI embodiments.
- Designed and trained bottle classifier VLMs (Qwen, SAM) in PyTorch for coordinate prediction and instance segmentation; moved inference from API to local edge for a 10x speedup.
- Built automated data tools and dashboards (React, FastAPI, MongoDB, AWS) and managed data operators for scalable training.
Terracotta
Sep 2024 - Jan 2025Hong Kong / Beijing (Hybrid)
Co-founder
- Led development of a cybersecurity SaaS product from 0 to market, achieving paying customers in 1 month.
- Established full-stack infrastructure (React, NodeJS, Python) and optimizing vendor costs.
- Secured top 10 position (out of 500+) in startup pitch competition and acceptance into MiraclePlus (formerly Y Combinator China).
Meituan
Jul 2024 - Aug 2024Beijing, China
Software Engineering (Machine Learning) Intern
- Optimized performance of multiple self-driving car machine learning models by up to 30%.
- Enhanced model efficiency using PyTorch's torch.compile and NSight Systems, leading to faster training times and reduced computational cost.
- Won Best Intern Team Award for outstanding performance in the internship program.
Sparticle
Sep 2021 - Sep 2022Tokyo, Japan
Front-End Development Intern
- Developed the front-end of a B2B web-based chat application using React in a multinational startup team.
- Led the English localization of Felo, a comprehensive workspace suite, translating over 2,000 phrases to enhance user accessibility.
Project Portfolio
Jittor RoFormer
Recreated RoFormer (Transformer with Rotary Position Embedding) in Jittor, a JIT deep learning framework in Python.
Five-Stage Pipeline CPU
Designed and implemented a five-stage pipelined RISC-V CPU in SystemVerilog, supporting RV32I instructions, memory-mapped I/O, serial communication, and interrupt handling.
Singularity EAM
Enterprise Asset Management (EAM) software developed in PHP, MySQL, and JavaScript for Singularity, a student-led software development team.