Six NVIDIA certifications
Tomek Brzozowski, Krzysiek Lenarski and Kuba Dyczko each passed two NVIDIA professional level exams: one on language models, one on agents. Plus a DGX station we now run models on in house.
There are two exams and they ask about entirely different things.
NCP-GENL takes language models from the engineering side. How to prepare the data, how to fine tune a model, how to spread training across several cards, how to fit inference into GPU memory and how to keep all of it running afterwards. NVIDIA assumes two to three years of hands on work with language models before this exam.
NCP-AAI goes one floor up, to agents: planning, memory, several agents working on one task, feeding the model knowledge that is not in its weights. Separately, questions about measurement, about the human in the loop and about safety guardrails — the part that separates a demo from something you can leave switched on.
Why NVIDIA when we live in SAP
Because GenAI stopped fitting inside SAP. Until now our language models worked next to the systems we already run. The NVIDIA path leads elsewhere: to Blueprints, the workflows and code samples NVIDIA publishes for building AI applications, and to deployments closer to the hardware, where what counts is card memory and response time rather than an integration pattern.
Then came the hardware. Yellow Cloud bought an NVIDIA DGX station and that is where we now work with language models in house: the model sits on our own machine, the data never leaves it, and we can take the thing apart and put it back together. Reading about fine tuning and GPU inference is one thing; doing it on your own card is another.
That is different work from integrating with SAP and it needs different things. Hence the exams and hence the hardware, rather than an introductory course.
Three people went through the same path, so we share a bar and a vocabulary. When someone here says “Agent Tomek”, nobody is thinking about politics.
A certification is not a project and we are not pretending otherwise. We run GenAI projects in parallel — we will write about them once they are finished, the way we write about everything else.
Sources
- Generative AI LLMs professional certification, NVIDIA
- Agentic AI professional certification, NVIDIA
- NVIDIA AI Blueprints, NVIDIA