~/blog / topic / ki
AI, LLMs & RAG in practice
What I learned building my own AI tools: RAG with citations, local models, agents with guardrails and how to make changes verifiable.
Halloween: the token monster prefers video
A Halloween still or a short AI video? One concrete example: 1,120 versus 28,960 output tokens, and why that ratio is not the whole cost story.
An AI assistant for your own website: chat, voice calls and appointments, hosted on my own Mac
How a local AI phone receptionist became an assistant for my website: widget, nginx proxy, voice on my Mac, appointment booking, privacy.
Topics: Nextcloud DNS, mail & certificates Linux troubleshooting Security & patches