Agent systems
Turn model capability into orchestrated, interruptible, shippable task loops
Multi-agent collaborationTool use & MCPTask state machinesHuman-in-the-loop
Hello, I'm
AI Kosei
Builder of AI agents & knowledge systems
Ship production AI with agent, RAG, and evaluation loops
Start from the scenario: orchestrate agents, govern knowledge, and absorb uncertainty with evals and guardrails
AI Native capabilities
Agent, knowledge, and evaluation skills for shippable systems
Knowledge & RAG
Answers that are grounded and permission-aware—not a black box
Multi-source retrievalAccess isolationRecall evaluationLiving knowledge updates
Evals & guardrails
Absorb model variance with metrics and constraints for safe launch
Offline / online evalsHallucination & privilege guardsCost & latency controlObservability & replay
AI-native architecture
Design platforms around domain boundaries—not stacked models and middleware
Scenario landingDomain modelingPlatform reuseLong-term evolution
Featured works
Public projects and artifacts
1 more require authorized access
Repo公开
AI克塞号站点（Astro）
个人门户与履历展示站点，Vercel 部署，GitHub OAuth 白名单访问。
Recent research
Engineering notes and retrospectives
