Repainting the world with cutting-edge technology
AI Articles
Lessons learned by actually breaking things in a high-capacity GPU environment — serialized on note (Japanese).
Profile
Shogo Okamuro
CEO / Technology Entrepreneur
Technology cannot change society if it stays in a lab or a demo. We go into the field, identify problems, build, test, and improve. At Nogeass, we use AI, software, and our own GPU infrastructure to tackle the difficult problems that businesses and local governments have struggled to solve.
I am committed to remaining an engineer who personally builds solutions and validates them in practice, rather than a consultant who only proposes ideas. That is how we approach making things.
What only Nogeass can do for you
Production-grade local LLM serving with vLLM
Built the full path from Ollama prototyping to vLLM API serving single-handedly. Can talk quantization, throughput, and the real operating cost of 72B-class models in numbers.
Self-hosted image & video generation pipelines
Goes beyond calling generative AI through APIs — runs the models locally, assembles them into pipelines, and operates them in production.
Real-time VLM agent experiments
Streams game frames to a VLM every second and automates state tracking, decision-making, and controller input. Battle-tested insight: state management matters more than image understanding.
AI assistants designed to know when to stay silent
Prototyping a JARVIS-style always-on assistant. Prioritizes judging when not to interrupt over speaking ability — an AI that helps without being asked.
Search and permission design that skips "default RAG"
Uses SQL, full-text search, vector search, and direct reference for the right jobs. Multi-tenant design that settles permissions in PostgreSQL and searches with OpenSearch.
Claude Code & LLM API design that avoids tech debt
Splitting CLAUDE.md and Skills, six-phase API pipelines, and cost designs that preprocess locally and reserve external APIs for the hard decisions.
Technical Advisory
From building AI infrastructure and judging the ROI of local LLM adoption to reviewing RAG/search architectures and Claude Code workflows — support from someone who has actually run it all.
Get in touch実績・体制
確かな技術力と、幅広い実績
Web3・AI等の先端領域を舞台にしたクリエイター集団。スタートアップ支援、大企業（コニカミノルタ等）の技術顧問、著名人（のん）とのNFTコミュニティ構築。シンガポール拠点でグローバルに展開しています。
Talks & Community Contributions
Slides from talks and seminars on putting AI to work. (Slides in Japanese)