Machine learning engineer / Engineering Manager @ Elith
I work across AI projects involving LLMs, computer vision, speech AI, and 3D/CAD. This page is the public index for selected work, writing, talks, and contact.
Current focus
I am an Engineering Manager at Elith, working across project execution, technical selection, design and implementation, code review, and team development improvements. Publicly, I focus on Kaggle, GPU Mode, AtCoder heuristics, research, and open-source AI evaluation. Employer- or client-specific information is intentionally not represented here.
Selected work
Self-taught in machine learning and reached Kaggle Competitions Master after two months of intensive competition participation. 1 gold, 6 silver, 3 bronze, peak world ranking 157, with support from the shared knowledge of the community.
Provisional GPU Mode Sprint Competition results include grayscale_v2: A100 3rd / B200 4th, and vectorsum_v2: A100 2nd, with a focus on practical kernel-level performance work. grayscale_v2 vectorsum_v2
Peak AtCoder Heuristic Contest rating: 1784. Profile
Research and evaluation work around reliable AI systems, safety cases, and measurable behavior under real constraints.
Contributions to AI vulnerability and evaluation tooling, with public work separated from confidential employer and client projects.
Technology
PyTorch, BERT, LoRA Fine-tuning, OCR, AI Safety, AI Red Teaming, Prompt Engineering
FastAPI, Next.js, React
Python, C++, TypeScript, JavaScript, Dart / Flutter
GCP, Docker, Firebase, Elastic Cloud
Writing and talks
Koki Inoue, Naoya Takashima, Hayato Fujihara, Shuya Higuchi, Kota Shimomura, Ryuta Shimogauchi, Takayoshi Yamashita. ICLR 2026 Workshop ICBINB.
Planned public notes on contest-driven low-level optimization.
Planned material on evaluation, red teaming, and implementation evidence.