Skip to content

Concepts — The Foundations

Everything else in Bee builds on these ideas. Understand them once and the rest of AI engineering clicks into place.

Overview

You can build a lot by treating an LLM as a magic box. But the moment something goes wrong — a weird truncation, a surprising cost, a model that "forgets" — you need a mental model of what's actually happening. This section gives you that model, from first principles, without requiring heavy math.

What you'll learn

flowchart LR
    T[Tokenization] --> H[How LLMs Work]
    H --> TR[The Transformer]
    H --> E[Embeddings]
    style H fill:#F5A623,stroke:#c77d00,color:#000

  • How LLMs Work

    Next-token prediction, training, and why it produces such general capabilities. Start here.

  • Tokenization

    Models don't see characters or words — they see tokens. This explains cost, limits, and odd behavior.

  • The Transformer

    The architecture behind every modern LLM, explained through the intuition of attention.

  • Embeddings

    Turning meaning into numbers — the foundation of search, RAG, and similarity.

Learning Objectives

By the end of this section you will be able to:

  • Explain what an LLM is doing when it generates text.
  • Reason about token limits and cost.
  • Describe the transformer's attention mechanism in plain language.
  • Explain what embeddings are and why they power search and RAG.

Suggested order

If you're new, read them in the order above. If you're here for RAG or agents, at minimum read How LLMs Work and Embeddings.

Tip

This is a 🟢 Beginner section — it assumes general programming knowledge but no machine learning background. Math is offered as optional depth, never as a prerequisite.