Amaete Build

Build AI from scratch.In 3D.

Over 650 hands-on lessons that open up how AI really works, from a single neuron to agents that use tools. Turn it, break it, fix it. Told from Africa, for everyone.

Start the first lessonSee every lesson

How numbers learn

Vectors, slopes, real fares and how blame flows back.

Seeing and reading

Filters that find edges. Words as tokens and points.

Inside a transformer

Attention, many heads, word order and the next word.

Models that use tools

Look it up, call a tool, loop, and plug in with MCP.

Build it, prove it

Build a database from nothing, then prep for the exams.

Scroll to open the cube

Turn it. Break it. Fix it.

A small machine in every lesson. Free, nothing to install, narrated.

Turn it

Drag to look around the machine. Every part is labelled, and the labels stay calm while you turn and zoom.

Break it

Each lesson has switches that break it on purpose: too big a step, a missing chain rule, two tools claiming one job. It tells you what went wrong.

Fix it

Then put it right and watch it work. Three quick questions at the end check that it stuck.

Roll a ball downhill.

How every model learns: one step downhill at a time.

Drag the slider and watch every step change. Tap the slope to drop the ball somewhere else.

Step 0: rolling downhill.
Open the full lesson

Told from Africa, for everyone.

Examples from across Africa, prices in your own currency.

Okada and boda boda faresJollof or nyama choma?Ajo, susu and chama savingsWill it rain in Accra?Matatu and trotro routesPuff-puff or mandazi?Doro wat and ta’ameyaOkada and boda boda faresJollof or nyama choma?Ajo, susu and chama savingsWill it rain in Accra?Matatu and trotro routesPuff-puff or mandazi?Doro wat and ta’ameya
Prices in naira, cedis, shillings, rand or CFAA mobile-money help deskIs the network down in Kisumu?The weather in MombasaTwo narrators to choose fromMakroud and mandaziTurn it, break it, fix itPrices in naira, cedis, shillings, rand or CFAA mobile-money help deskIs the network down in Kisumu?The weather in MombasaTwo narrators to choose fromMakroud and mandaziTurn it, break it, fix it

650 lessons. Start anywhere.

Twelve tracks, from one neuron to agents. About ten minutes each.

Numbers that learnVectors: how alike are two things?Turn jollof, suya, nyama choma and makroud into arrows.Numbers that learnGradient descentRoll a ball down a hill of mistakes.Machine learning basicsLinear regressionGuess a motorbike-taxi fare from the distance.Neural networksThe perceptronTeach one artificial neuron to tell which plantains are ready for dodo.Neural networksBackpropagationWatch a small neural network learn.Neural networksActivation functionsThe small bend inside every neuron.SeeingConvolutionsSlide a 3×3 window over an adire pattern built from pillars.LanguageTokens: how AI reads textWatch a tokenizer chop sentences into pieces on a conveyor.LanguageWord embeddingsFly through a space where every word is a point.TransformersSelf-attentionBoard a danfo (a matatu in Nairobi, a trotro in Accra) where every word is a passenger.TransformersMulti-head attentionFour attention heads, stacked like glass shelves over one sentence.TransformersPositional encoding"Lion chases goat" and "Goat chases lion" use the same words.Making an LLMNext word and temperatureA tiny language model offers its guesses for the next word as towers.Making an LLMQuantization: shrinking a modelSqueeze a model to fit on a phone.Using LLMsRAG: look it up firstA co-op's notes sit on a shelf.Using LLMsFunction callingWatch a model ask for help.AgentsThe agent loopFollow a dispatch rider who is really an AI agent.AgentsMCP: plugging in toolsSee how an AI app plugs into files, calendars and weather through one standard socket.Build it yourselfBuild your own databaseKeep a savings group's book in a tiny database you can see inside.Exam prepExam prep: designing toolsAn agent picks tools by reading their names and descriptions, nothing else.