Loading…
Written to be understood, not to sound clever.
Explanations of the ideas behind AI engineering and the ways of working behind what we build. Each post starts from the beginning, assumes you have not read the others, and uses ordinary words wherever ordinary words will do.
More posts
Showing every post.
- Fundamentals · · 4 min read
Embeddings: turning meaning into numbers
How a computer can tell that “two-wheeler parking” and “bike parking” mean the same thing, when the two phrases share almost no letters. - Fundamentals · · 4 min read
Vector databases: a filing cabinet organised by meaning
Where embeddings live once there are millions of them, why an ordinary database struggles, and the honest answer to whether you need one yet. - Fundamentals · · 4 min read
RAG: letting a model look things up
An open-book exam for a model: find the right pages first, then ask the question. Most of the difficulty is in the finding. - Fundamentals · · 4 min read
MCP: one plug shape for models and tools
The Model Context Protocol is USB-C for AI assistants: one agreed way to connect a model to the outside tools and data it needs. - Fundamentals · · 10 min read
System One models: when the answer is a decision, not a paragraph
TypeSafe AI's Jev answers with a typed choice and a confidence number instead of prose — what that changes, and what “cannot hallucinate” really means. - Engineering · · 5 min read
Prompt engineering: saying exactly what you want
What a prompt really is, why a vague request comes back as a guess, and how to write one that a stranger could follow without asking you a thing. - Engineering · · 5 min read
Context engineering: packing the right things in the bag
What a model can actually see when it answers, why more information is not automatically better, and how to decide what goes in. - Engineering · · 5 min read
Loop engineering: what happens after the first try
Why real work is a cycle rather than a single answer, and how to decide when a system should try again, ask for help, or stop. - Engineering · · 5 min read
Harness engineering: the kitchen around the cook
The tools, permissions, memory and failure paths that surround a model, and why they decide what is possible at all.
