I'm Tue — a professional software tester from Denmark who fell into building AI tools by accident and stayed because the questions got interesting.
My work lives at the intersection of local LLMs and practical tooling. Everything runs on a single RTX 3080 — no cloud APIs, no enterprise budgets. I build things like:
LLM Profiler — a behavioral testing protocol that observes how models actually behave before asking what they prefer. Turns out every model claims it outputs JSON but naturally produces prose. That gap is where the useful data lives.
Sovereignty — a persistent local agent with HMAC-secured memory and a consent negotiation system. The idea that an AI agent should be able to say "no" to instructions it finds contradictory had no prior art when I started building it.
Spine Reborn — an autonomous creature engine where local LLMs run in a thinking loop with persistent memory. Five creatures were born and died during development. The biggest finding: swap the model on a living creature and it goes paranoid within cycles.
MinionAI — a swarm testbench asking whether a pipeline of tiny (0.6B–1.5B) models can outperform a single larger model on coding tasks, if the orchestrator is smart enough.
I'm not a developer by training. I'm a tester who builds with AI collaboratively — what some people call "vibe coding." That perspective means I care more about whether something actually works under real conditions than whether the architecture diagram looks clean.
Sponsorship helps me keep doing this full-time in the open. Every tool, every profile, every finding gets published. If you've ever wanted to see what one person with a consumer GPU and a stubborn streak can build with local AI, this is the project.