
How an LLM works: from your question to the first word
Tokens, attention, KV cache: opening the black box, diagram by diagram, to design better AI projects and keep their costs under control.
29 October 2026
Articles written by the team, about the questions we run into on engagements: data, AI, cloud. Without jargon whenever possible.

Tokens, attention, KV cache: opening the black box, diagram by diagram, to design better AI projects and keep their costs under control.

An AI agent can explore a situation, use tools and adapt its actions. A workflow can organise a reliable, repeatable process. The right choice depends on the problem, your constraints and the benefit you can actually measure.
A project, a question, a doubt about an architecture: email us or book a slot. The first conversation is for understanding, not selling.