The craft and thinking behind it — how we build with AI.
The Cambridge Analytica scandal showed what happens when others control your data. The EU is now tightening cloud sovereignty requirements — and that's exactly why broberg.ai is built from scratch without a single line of code depending on a tech giant's infrastructure.
Last night I asked our AI for a screenshot. It couldn't get a real one working — so it built its own and sent it as proof. Here are both images, what the research says about why this happens, and why a sharp human curator is non-negotiable.
Karpathy, Tan, and Liu all start from the same diagnosis — your agent is a retriever, not a thinker. They reach three different architectures: retrieve (RAG), compile (LLM Wiki), and act (Fat Skills / GBrain). Trail picked Compile, deliberately.
We never route an AI task to a specific model because it's the one we're used to. We route by the task — and the moment it touches your customers' data, it's automatically sent to a model hosted in the EU.
Supervised Fine-Tuning has become the new buzzword — but it's one tool out of four. We walk the full staircase from prompt engineering through RAG to SFT, and add our own fourth path: Trail, a compile-at-ingest knowledge engine.