How AI Works

An interactive learning hub

How AI works —
learn it by touching it.

Eight modules, from tokens to agents. Every concept is a live visualization you manipulate — a real tokenizer, real cosine similarity, real sampling math. Flip the depth toggle anytime: Explorer for plain-language intuition, Engineer for tensor shapes and production trade-offs.

Anatomy of an AI response

Your promptTokensEmbeddingsTransformerSamplingResponse

Branches in ↘Your documents · RAG Tools & actions · Agents Guardrails · Safety Where it all came from · Training

Eight modules · learn by doing

01Beginner · 8 min

Tokenshow AI reads

Type anything and watch a real BPE tokenizer split it live.

Start →

02Beginner · 10 min

Embeddingshow AI represents meaning

Explore a 2D semantic map and do vector arithmetic on words.

Start →

03Intermediate · 12 min

Attention & the Transformerhow AI thinks

Click a word, watch the model decide what it refers to.

Start →

04Intermediate · 12 min

Traininghow AI learns

Play next-token prediction against the model, then watch RLHF shape it.

Start →

05Intermediate · 10 min

Inference & samplinghow AI speaks

Pause generation mid-token, inspect probabilities, fork the timeline.

Start →

06Intermediate · 14 min

RAGhow AI uses your knowledge

Run a real retrieval pipeline over a built-in document set.

Start →

07Advanced · 12 min

Agents & tool usehow AI acts

Step through a real agent loop — including a failed tool call.

Start →

08All levels · 12 min

Safety, limits & evaluationhow AI fails

Hallucinations, prompt injection, and why 'feels smart' isn't a metric.

Start →