>_ terminal 日本語
Taiki Metoki

Taiki Metoki

I train models for drug discovery and language 🧬🤖

Hugging Face
M.S. student, Yokohama City University
2026.06 –
I am an intern at Matsuo Institute, where I mainly work on research related to LLMs.
2026.05 –
I am a research assistant at NII LLMC / LLM-jp, working on reinforcement learning and LLM post-training. I first joined in October 2025 as a technical assistant.
2026.04 –
I am an M.S. student at the Graduate School of Medical Life Science, Yokohama City University (Structural & Cellular Science Lab). Through the cooperative graduate school program I carry out my research at AIST, working on AI drug discovery and bioinformatics.
2025.09 –
I am an intern at Elith, working on GENFLUX.
2025.07 –
I am a technical trainee on the Machine Learning Research Team at AIST AIRC (Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology), working on bioinformatics and AI for drug discovery.
2025.05 – 2026.05
I was a data scientist intern at Cubec, developing medical LLMs.
2025.02 – 03
I was a research intern at NAIST, estimating lumbar bone mineral density from X-ray images.
2024.12 – 2025.03
I took part in LLMATCH Season 1, exploring how LLMs can be used in drug discovery research.
2024.05 – 08
I was an intern at Hips, working on customer data analysis.
2022.04 – 2026.03
I earned my B.Eng. at the Faculty of Textile Science and Technology, Shinshu University. In the Shimada Lab I worked on deep learning for chemical reaction prediction.
bio
Taiki Metoki is an M.S. student working on AI drug discovery, LLM post-training, and AI safety. As an undergraduate he studied deep learning for chemical reaction prediction, and he now researches AI drug discovery and bioinformatics at Yokohama City University. Alongside his degree he works with AIST, NII LLMC / LLM-jp, and Matsuo Institute, covering bioinformatics, medical imaging, and reinforcement learning from both the research and the engineering side. He is a Kaggle Competition Expert with two silver medals.


competitions
🥈 SILVER
19 / 1391
Vesuvius Challenge — Surface Detection
Model development with nnU-Net, plus post-processing. 2026.02
🥈 SILVER
133 / 3802
CSIRO — Image2Biomass
Led DINOv3-based modeling; ~50 experiments for model selection and subtask validation. 2025.12–2026.01
🧠 SOLUTION
2.5D baseline
RSNA — Intracranial Aneurysm Detection
Built and open-sourced a 2.5D baseline. 2025.09–10
🏅 RUNNER-UP TEAM
SFT / GRPO
Matsuo Lab LLM Development Competition 2025
Led SFT/GRPO training of Qwen3 models in Phase 1; handled infrastructure in Phase 2. 2025.07–09
research
AI Drug Discovery — bioinformatics and machine learning for drug discovery.
LLM Post-training — reinforcement learning, reasoning, and domain adaptation.
AI Safety — trustworthy LLM behavior and how to evaluate it.
Medical Imaging — detection, segmentation, and diagnostic support models.
Scientific ML — chemical reaction modeling, Neural ODEs, and mathematical models.
publications
Toward Learning Amino Acid Substitution Matrices via Inverse Reinforcement Learning
The 26th Annual Meeting of the Protein Science Society of Japan (PSSJ), 2026 (poster)
Taiki Metoki (Yokohama City University, National Institute of Advanced Industrial Science and Technology (AIST)), Fangzhou Lin (Texas A&M University, Tohoku University), Kazunori D Yamada (Tohoku University, Ochanomizu University), Kentaro Tomii (AIST, Yokohama City University)
Extending Chemical Reaction Neural Networks to Heterogeneous Catalytic Systems and Improving Prediction Accuracy
Society of Computer Chemistry, Japan — 2025 Autumn Annual Meeting in Gifu (poster)
Taiki Metoki, Natsuki Yokosuka, Iori Shimada (Shinshu University)
misc
My usual tools are Python / Julia / PyTorch / Docker / Git / HPC.

I am grateful to the following foundations for supporting my studies.