Skip to content

Learning Paths 🗺️

A pile of articles isn't a curriculum. Learning paths sequence Bee's content so each concept builds on the one before it. Pick the path that matches where you are — and where you want to go.

How to use a path

Follow it top to bottom. Each step links to a Bee page. Do the exercises — they're where the learning sticks. Paths reference content across the whole hive, including sections still being written (marked 🚧); those links light up as content lands.

Choose your path

  • 🟢 AI Engineer — Fundamentals

    You can code, but LLMs are new to you. Build a correct mental model of how these systems actually work.

  • 🟡 Building with LLMs

    You've made API calls. Turn them into reliable features: structured output, tools, and good prompts.

  • 🟡 RAG Specialist

    You need models to use your data. Master retrieval end-to-end.

  • 🔴 Agent Engineer

    You want autonomous, tool-using systems. Planning, memory, and multi-agent design.

  • 🔴 Production & MLOps

    You need to ship. Evaluation, safety, deployment, and monitoring.

🟢 Path 1 — AI Engineer: Fundamentals

Goal: understand what an LLM is doing well enough to reason about its behavior and cost.

  1. Setup & Prerequisites
  2. Your First LLM Call
  3. How LLMs Work
  4. Tokenization — why the model "sees" tokens, not characters
  5. The Transformer — attention, intuitively
  6. Embeddings — turning meaning into numbers
  7. Prompt Engineering — your first real skill

You'll be able to: explain next-token prediction, estimate cost, and write clear prompts.

🟡 Path 2 — Building with LLMs

Goal: ship reliable LLM features, not demos.

  1. Prompt Engineering
  2. System Prompts
  3. Structured Outputs — get JSON you can trust
  4. Function & Tool Calling — let the model do things
  5. Evaluation 🚧 — know if your changes help or hurt
  6. Security 🚧 — prompt injection and guardrails

You'll be able to: build a feature with validated output, tool use, and basic evals.

🟡 Path 3 — RAG Specialist

Goal: make a model answer accurately from your documents, with citations.

  1. Embeddings
  2. RAG Overview
  3. Chunking — the single biggest lever on RAG quality
  4. Vector Databases
  5. Hybrid Search & Reranking
  6. Evaluating RAG

You'll be able to: design, build, and evaluate a production RAG pipeline.

🔴 Path 4 — Agent Engineer

Goal: build systems that plan, use tools, and act over multiple steps.

  1. Function & Tool Calling
  2. Agent Fundamentals
  3. Memory
  4. Multi-Agent Systems
  5. Model Context Protocol (MCP)
  6. Evaluation 🚧 — evaluating non-deterministic agents

You'll be able to: design an agent loop, give it memory and tools, and know its failure modes.

🔴 Path 5 — Production & MLOps

Goal: take an AI feature from laptop to reliable, observable production.

  1. Evaluation 🚧
  2. Security 🚧
  3. Deployment 🚧 — Docker, serving, scaling
  4. MLOps 🚧 — CI/CD, observability, cost monitoring

You'll be able to: ship an AI service with tests, guardrails, monitoring, and cost controls.

Want a path we don't have yet?

Request one or design it yourself — learning paths are one of the most valuable things you can contribute.