Yichong Zhao / Eric Zhao (赵奕冲) is an AI engineer at Preferred Networks, a Tokyo-based AI unicorn, in its AI Products & Solutions division — where he builds LLM-based agents for operations-research problems and, more broadly, for the kind of complex, open-ended problems found in the real world. He also works on the post-training of nucleic-acid sequence language models within a JST CREST project.
Research interests
AI Products & Solutions · Tokyo, Japan
Tokyo, Japan
University of Tokyo (Matsuo Lab)–originated AI startup
Recommender systems & MLOps
Web development & web crawling
University of Tokyo · Graduate School of Frontier Sciences
Advisor: Susumu Goto
Waseda University · School of Advanced Science and Engineering
Advisor: Michiaki Hamada
Can Frontier LLMs Replace Annotators in Biomedical Text Mining? Analyzing Challenges and Exploring Solutions
arXiv preprint, arXiv:2503.03261, 2025
Benchmarked frontier LLMs against BERT-family models on biomedical text mining, analyzing why LLMs underperform and the domain-specific constraints that limit them.
Enhancing Biological Text Mining with LLMs: Systematic Prompt Engineering and Data Augmentation
Asia-Pacific Bioinformatics Joint Conference (APBJC 2024)
Development of an RNA-Generative AI Model via Large-Scale Pre-training (大規模事前学習モデルによるRNA生成AIモデルの開発) JMAI Research Award
Japanese Association for Medical AI (JMAI 2024)
Programmed a Unitree G1 humanoid robot to perform traditional Japanese mochi-pounding (mochitsuki) — awarded 1st place for its completeness and originality at this international physical-AI hackathon.
Trained an 8×7B large language model from scratch within Japan's GENIAC initiative (supported by METI), collaborating with a Matsuo Lab team. I was primarily responsible for the model-training phase.
weblab-GENIAC on Hugging FaceBuilt an AI mock-interview app that won multiple awards at the 100 Program hackathon and drew interest from an incubator.