A new way of learning
Animated tutorials with labs you run yourself.
readcutaway.com
- Watching endless tutorials
- Reading endless blogs
- Memorizing definitions
- Building a demo that works once
AI engineering is making the call.
The questions that come up after the demo, when there are real users, a budget and an on-call rota. Every volume answers four of them: why, how would I design it, how would I debug it, and how would I take it to production.
Pick where you are.
Guided paths, each read in order, from your first idea to the decisions your job asks of you. Pick the one that fits.
- AI Basics
- How AI Works
- Talking to a Model
- Giving a Model Your Own Data
- and more…
- Inside the Model
- Inference Performance
- Adapting Models
- Retrieval at Depth
- and more…
- What You're Funding
- The Money
- The Risk
- Quality and Trust
- and more…
Every idea, animated.
One page per idea, from tokens and the KV cache to prompt injection. Watch each mechanism move, at your own pace.
Start here: how large language models work
FreeReuse past attention work
FreeMeaning, written down as coordinates
FreeReorder candidates by relevance
FreeThe model asks; your code does it
FreeUntrusted text starts giving the model orders
FreeGo deep. Then build it.
Animated, step by step, from how the mechanism works to when to stop. Every tutorial comes with labs you run on your own machine.
Foundations
Deciding what travels with every request, and what gets looked up.
Coming soonStep 1 freeBuilding a search that follows links instead of matching words.
Coming soonStep 1 freeBuilding the code around an agent, and measuring what each layer buys.
Coming soonStep 1 freeChoosing how to arrange an agent system: one agent, a pipeline, a router, or more.
Coming soonStep 1 freeWhat to retrieve, how to find it, how much to hand over, and when to return nothing.
Coming soonStep 1 freeThe go or no-go before anything is built: what a wrong answer costs, what would do the job without a model, and what working has to mean on day one.
Coming soonStep 1 freeReliability
When to give an agent another attempt, and when to add another agent.
Coming soonStep 1 freeWhen an agent should remember, and when it should look it up again.
Coming soonStep 1 freeWhen an agent writes its own rules, and how to tell if the work got better.
Coming soonStep 1 freeWhat users see when the model provider is slow, limited or down, and which of the usual fixes make it worse.
Coming soonStep 1 freeProduction
How to tell whether your eval's number means anything.
Coming soonStep 1 freeA guardrail’s value is set by where it sits, not by how good it is.
Coming soonStep 1 freeWhich cost reductions cannot cost you an answer, in what order, and when to stop.
Coming soonStep 1 freeWhat to record, how to release a change safely, how you notice it went wrong, and how you find out why.
Coming soonStep 1 freeWhich model is good enough for your task, how to prove it, what it really costs, and how to survive the next model change.
Coming soonStep 1 freeThen run it yourself.
Every tutorial ships with labs: small, plain-Python programs that reproduce the mechanism on your machine, in under a second, with no API key. This is real output from the Eval Engineering lab.
The questions you'll be asked.
Your team ships with agents now. Short, standalone volumes, every figure marked as cited, modeled or an exhibit, so you can repeat it in a meeting.
“Why aren't we shipping faster with agents?”
Free · Read now →“My team says 94%. Should I believe it?”
Coming soonIn the bundle“What do I say when an agent gets it wrong?”
Coming soonIn the bundle“Are our metrics still telling us anything?”
Coming soonIn the bundle“Which work should we hand to agents?”
Coming soonIn the bundle“If juniors skip the grunt work, how do they grow?”
Coming soonIn the bundle“What are we losing without noticing?”
Coming soonIn the bundle