Skip to content

Playlist - AI engineer, 30 minutes

Six concept pages in order, total ~30 minutes. By the end you have working mental models for: how LLMs predict the next token, how you give them access to your data, how you give them tools, and how you wire that into a working agent.

The reads, in order

  1. LLM basics (~5 min) What an LLM is, tokenization, the next-token-prediction loop. Foundation for everything else.

  2. Embeddings and vector search (~5 min) How meaning becomes a vector and how nearest-neighbor search lets you "find similar things." Required before RAG makes sense.

  3. RAG explained (~5 min) The workhorse pattern: retrieve relevant chunks → put them in the prompt → answer. Why RAG beats fine-tuning for "give the model access to my data."

  4. Tool use and function calling (~5 min) How the model returns structured JSON that your code dispatches as a tool call. The mechanism under every agent.

  5. MCP explained (~3 min) The standard tool interface that decouples agents from server implementations. Why this matters in 2026.

  6. Agentic loops (~7 min) The runtime that turns tools + a model into an agent: take output, execute tools, feed back, repeat. Where most production engineering lives.

What you can do after this playlist

  • Articulate the difference between RAG, fine-tuning, and prompting, and pick the right tool for a problem.
  • Sketch the architecture of a Claude / GPT agent that reads files, queries a database, and answers natural-language questions.
  • Discuss why MCP exists and what problem it solves.
  • Recognize when to reach for an agent vs a single LLM call.

Next steps after this 30-minute foundation

If you want to build: - Build a RAG pipeline - end-to-end with pgvector + Claude (~30 min) - Build a Claude agent with MCP - tool-using agent with custom MCP server (~30 min)

If you want to go deeper: - Evals for LLMs - the regression test for LLM apps - Context windows and management - the operational reality of agent context - Prompt caching - cost optimization for stable-prefix prompts - Inference servers - if you're going to self-host

If you want a cert: - AI/ML systems topic index - all relevant certs grouped by tier - LLMs and GenAI topic index