“Capstone: Anatomy of an LLM App”, the finale of the How LLMs Work track. One customer’s message travels the whole stack, and every board in the track turns out to be one hop on that journey.
Before you start. You need the returns desk from Agents & Tool Use, where the clerk who can only talk is the large language model itself, the printed forms are its tools (ordinary functions in your code, like lookup_order, described to the model so it can ask for one by name), and the back office that fills a form in is your code running that function. You also need one fact from Inference (sampling, context, KV cache, hallucination): the model remembers nothing between sentences, so somebody has to copy the whole conversation out again every single time.
Continue reading with a plan
The first three lessons of every track are free. This lesson is included with a plan, which unlocks every lesson, board, video, lab and paper for one price.
