▍ Local inference · agentic systems · Tartu
This is where I write about that work — what I'm testing, what breaks, and what I find. Diary-shaped, evidence-first. No conclusions I haven't run.
An offhand conversation — "what if we had a self-hosted NotebookLM?" — grew into a research system over ~8 million chunks of Estonian law: six indexes I built, hybrid retrieval, live calls into the state registers, and a multi-agent Tribunal that cross-examines every answer before it reaches you. It solved real cases for real people — or honestly said it couldn't. The story, and why I paused it.
Read the story →Wiring a locally-hosted GLM into the WhatsApp group i share with a friend — then giving it web search, image generation, proxy 'eyes', and the ability to watch YouTube. A 24-hour build, told in chat messages.
Read the finding →Taking the WhatsApp companion further: a fleet-ops co-pilot to run a 6-node GPU homelab from my phone — two AI operators sharing one seat, gated by a trust model. The design so far, plus a general blueprint anyone can use.
Read the finding →Summer 2025, one GPU bought on a hunch, a €9,000 receipt and the regret that came with it — and the project that turned it all into four Blackwells. The build, in photos.
Read the origin story →
homelab