What this is

I'm Damir Radulić. I run RiNET. We build autonomous AI infrastructure for clients who'd rather ship a working system than receive a glossy slide deck.

This blog is the field-notes channel. Topics that will show up here:

• Sovereign stacks: how to run end-to-end inference, embeddings, vector search and graph reasoning on your own EU iron — and when not to bother.
• Agent swarms: long-running planners, sub-agents, tool routers. One brain, many personas. Hard guardrails on facts.
• LoRA & fine-tuning: when adapters beat prompting; when they don't; how to evaluate honestly without fooling yourself.
• Postmortems from generalized client work — public-sector clients in SE Europe, energy-sector pilots, civic-intelligence platforms — with names changed and lessons kept.

What it is not

Not "thought leadership". Not 5-things-I-learned-from-GPT-4 carousels. Not takes on yesterday's model release. If you want timeline-shaped opinions, the rest of the internet is well stocked.

Why bother

Because most of what's written about AI right now is downstream of demos. The infrastructure that decides whether an AI product actually ships rarely makes it into the blog post. The interesting failures don't trend. The boring decisions — connection pooling, idempotent embed jobs, the third backup of a fine-tune checkpoint — never do.

Those are the parts I find worth writing about. If you find them worth reading, good. The RSS feed is right there.

— Damir