Recent Lessons & Consultations
Rebuilding the attention mechanism from one premise: a matrix multiply is a projection.
How do Q, K, and V matrices actually work? Let's reframe the attention mechanism into something you can rediscover by intuition. One worked example runs through the whole lesson, mixing token rows of “the cat sat on the mat” until causal masking and next-token prediction fall out of the matrices.

Can an AI reason through a constrained engineering purchase?
Most ecommerce demos start with “find me shoes in size 8.” This one starts with a multilayer ceramic capacitor, the part that stops a production line when it goes missing. The case study builds an agent that works through electrical and manufacturing constraints to an actual order.
