YYYosuke Yoshimurainworldreadingnote.hashnode.dev·Sep 29 · 2 min readResearch Brief RAG Agent: Grounded Answers with Explicit AbstentionA local RAG work sample designed to show not only what the system answers, but when it should stop. The problem Knowledge applications need to answer questions across multiple documents while keeping 00
YYYosuke Yoshimurainworldreadingnote.hashnode.dev·Sep 29 · 3 min readA One-Line OSS Fix That Was Worth Merging: Keeping Evidence Claims HonestHow a +1/-1 documentation patch aligned a project’s known gaps with its actual evidence Small documentation changes are easy to underestimate. Sometimes the most useful change is not adding new eviden00
YYYosuke Yoshimurainworldreadingnote.hashnode.dev·Sep 22 · 4 min readThree Small Tools for Making AI Work More InspectableAI systems increasingly make claims, call tools, and change files or configurations. The difficult part is often not generation or execution itself. It is understanding what happened around the bounda00
YYYosuke Yoshimurainworldreadingnote.hashnode.dev·Sep 12 · 4 min readBefore You Paste It Into an External AI: Make the Unknowns VisibleExternal AI tools are useful when a technical task is stuck: explain a production error, review a diff, summarize an incident timeline, or turn a rough support note into a clearer next step. The diffi00
YYYosuke Yoshimurainworldreadingnote.hashnode.dev·Sep 8 · 5 min readBefore You Create Another Draft, Check What Already ExistsDuring one recent AI-assisted publishing session, the draft list was checked before the published list. A new draft was started for an article that was already public. The platform had not lost the ar00