Certaines informations sont présentées en anglais.
À propos de moi
I build small, reliable LLM systems: RAG over documents in Japanese and English, MCP servers that let Claude use your own tools and APIs, and multi-step agents with real verification loops. I focus on the parts that usually break: retrieval quality, idempotent writes, and agents that check their own output instead of failing silently. I work from a written spec and ship tested code. Recent public examples on GitHub (AgentGymLeader): japanese-rag-starter, mcp-server-starter, and agent-workflow.... Plus d’infos