LLMs & Generative AI
From tokenization to KV cache math — a self-paced track covering how LLMs work, how to build with them, and how to actually serve them on a GPU.
TopNotchNote
Notes that matter
foundations
What a transformer actually does, and how a raw next-token predictor becomes a model you can chat with.
prompting & practical use
The cheapest lever for steering a model, and the practical mechanics of getting reliable output and actions out of it.
embeddings & RAG
Grounding a model in facts it wasn't trained on, without touching its weights.
fine-tuning & adaptation
When updating the weights themselves is actually the right tool, and how to do it without a training cluster.
inference & serving
The GPU-level internals of running these models in production — where this track connects directly to the GPU Programming section.
agents & applications
Putting the pieces together into systems that act, and knowing whether any of it actually worked.