Post

I was validating models long before "Proof" had a name for it.
In my experience, most ML engineers already does a version of RAMP without calling it that. You retrieve and clean the data before training. You choose between classical ML and an LLM depending on what the task actually needs, not whichever is trendiest. You validate every prediction before it goes anywhere near a decision, because a model that's confidently wrong is worse than one that's slow.
What RAMP adds isn't the four ideas, ML engineers already live inside three of them. It's Agents, that's the one actually new to how I work. Delegating a workflow to something that plans and acts on its own, not just returns a prediction, is a different discipline than anything model validation ever prepared me for.

So when I ask myself if I'm AI-native, Retrieval, Models, and Proof don't really test me, that's just the job I already do. Agents is the one that actually does.

More on RAMP: aidoos.com/ramp-framework

#RAMP #RAMPFramework