no code · fully interactive · the whole stack
See AI Think.
A finishable course on how modern LLMs actually work. You build the intuition by operating the machine, not watching videos.
Predict first
Which word does the model reach for most? Tap your guess.
From AI buzzwords to systems literacy
Understand how LLMs, agents, RAG, evals, and tools actually work, so you can reason about modern AI products with confidence instead of collecting prompt tips.
- Before
- AI feels like a black box. You collect prompt tips and hope they transfer.
- After
- You reason from how it actually works: tokens, attention, RAG, agents, and evals, and make real calls about AI.
Updated for the 2026 stack
- RAG
- agents
- MCP
- computer use
- model routing
- retrieval engineering
- multimodal
- voice
- evals
- governance
What you'll be able to do
- Explain how a model turns words into numbers, then predicts the next one.
- Read attention: why the same word means different things in different sentences.
- Tell a real agent from a chatbot, and see where tools and permissions belong.
- Design RAG, and know when retrieval is the fix and when it is not.
- Choose which model to use, and what an eval should actually check.
- Spot prompt injection and the new ways AI systems fail.
Free, no account needed. About 5 minutes, ends with one clear win.
Built so the ideas stick
The same four moves on every lesson. Not videos to watch, a loop that makes the ideas land and stay.
Predict first
Commit a guess before anything moves. Being wrong on purpose is what makes the reveal land.
Touch the machine
Drag the vectors, sharpen the attention, turn the crank. The idea is a thing you operate, not a paragraph you read.
Recall later
Spaced questions bring each idea back right before you'd forget it. Recalling it is what locks it in.
Prove it in a lab
Boss labs hand you a whole act at once. You apply the machine, you don't just recognize it.
Pick a route by goal
Start with the Main Path, the essential route through how AI works, a finishable 34. The goal tracks and challenges are optional depth layered on top, reach for them any time.
I want the mental model
Start hereThe Main Path
The essential path. How AI actually works, end to end, from a word to attention, scaling, RAG, and the limits.
34 lessons · ~2.8 h
0/34 done
Start →I want the 2026 stack
2026Modern AI Systems
The 2026 stack, end to end: RAG and retrieval engineering, tools, agents, MCP, computer use, multimodal, voice, evals, model routing, and governance.
28 lessons · ~2.3 h
0/28 done
Start →I use AI at work
AI at Work
Drive AI tools well, context, agents, permissions, and habits that get more from every prompt. No code.
29 lessons · ~2.4 h
0/29 done
Start →I build AI features
AI Product & Trust
Ship and run trustworthy AI features, the pipeline, observability, evals, data, security, and honest UX.
25 lessons · ~2.1 h
0/25 done
Start →I want extra challenges
Challenges & Deep Dives
Off the main spine: end-of-act boss challenges, the capstone, and elective rooms for extra depth.
28 rooms · ~2.8 h
0/28 done
Explore →The course at a glance
The full arc, from a word to agents. The main path is the finishable spine; the rest is optional depth.
- 0The Big PictureSee it work first0/1
- 1RepresentationA word is a vector0/6
- 2Prediction & LearningHow it gets good0/5
- 3ArchitectureAttention & the transformer block0/9
- 4Operating & ScalingDecoding, context, scaling laws0/5
- 5A System in the WorldFrozen model → RAG → agents0/4
- 6Agents in PracticeDrive the tools well16 lessons
- 7LLMs in ProductionShip, watch, evaluate, improve0/1
- 8Physical SubstrateGPUs & datacenters0/1
- 9Elective RoomsOff the main arc0/2
Or just look one thing up
Not ready for a whole course? Every idea also has a plain-language explainer: the answer up front, what people get wrong, and a free interactive lesson at the end of it.
Why this exists
I built this because most AI explainers either stop at prompt tips or jump straight to the math. This sits in between: you learn how modern AI actually works by operating the real machinery, made visible and hands-on, so the understanding is yours, not borrowed.
It is one person's work, built in the open and actively maintained for the 2026 stack, with new lessons as the field moves. One purchase includes the updates, and the first lessons are free so you can judge the quality before paying.
Made byFernando Carril- First lessons free, no account
- One-time purchase, lifetime access
- Updates included as the stack changes
- 30-day money-back guarantee
Questions, answered
Updated for mid-2026No code, finishable, current, and low-risk to try.
Do I need to know how to code?
No. There's no code and no math homework. You operate visual machines: drag the vectors, turn the dials, run the agent loop. If you can use a web app, you can do every lesson.
Who is this for?
Curious professionals, operators, product and design people, founders, and anyone who keeps hearing about LLMs, agents, and RAG and wants to genuinely understand how they work, not just the buzzwords.
Who is it not for?
If you want to train models in PyTorch or ship production ML code, this isn't that course. It builds the mental models underneath, the why, so you can reason about AI systems. It is not a coding tutorial.
How long is it, and can I actually finish it?
It's built to be finishable: a main path of 34 lessons you can complete in a few focused sittings, plus optional deep-dive rooms and labs when you want more. Spaced-recall questions bring ideas back so they stick.
Is the content current?
Yes. It covers the mid-2026 stack: agents and tool loops, RAG and retrieval engineering, evals and LLM-as-judge, multimodal, model routing, computer-use agents, and AI governance.
Will it stay updated?
Yes. The field moves fast and the course is maintained. One purchase gives you the updates as the stack changes.
What if it's not for me?
Start with the free lessons, no account needed. Full access is a one-time purchase with lifetime access and a 30-day money-back guarantee, so trying it is low-risk.