One roadmap, two lanes
Learn LLMs: Beginner to Advanced
A step-by-step path from "what is an LLM" to building and evaluating AI systems — with every concept shown twice: a no-code walkthrough in ChatGPT, Claude or Gemini for non-developers, and a code example for developers.
Lane 1 · Non-developers
You use ChatGPT, Claude or Gemini — and want to use them well
Managers, analysts, teachers, founders, support, HR, legal, marketing. Every stage has a no-code walkthrough done entirely inside the apps: prompts you can paste, Projects and Gems you can build, agents you can delegate to, and a scorecard so you know when to trust the output.
Lane 2 · Developers
You write code — and want to build with LLMs properly
Every stage has a runnable example in Python, implemented for Gemini, Claude and OpenAI in a tab switcher: the first API call, structured output, a minimal RAG pipeline, function calling, a first agent, and an eval harness. It hands off to the Agent Workflows track at the end.
The six stages
Each one builds on the last. Beginner → Intermediate → Advanced.
- 1Open →Beginner2–3 hours
What an LLM Actually Is (and Isn’t)
Before prompting tricks or code, get the mental model right: an LLM predicts text, works inside a limited window, and can be confidently wrong. Everything else follows from those three facts.
No-code: See the strengths and the limits in 20 minutes
Code: The one API call under everything. Install the SDK, set the API key as an environment variable, run.
- 2Open →Beginner3–4 hours
Prompting That Works: Role, Task, Context, Format
Most disappointing answers come from vague requests. Learn the four-part prompt, the two techniques that matter most (examples and step-by-step), and how to iterate instead of restarting.
No-code: Turn a messy meeting transcript into decisions, owners and deadlines
Code: System prompt + structured (JSON-schema) output. The model returns parsable data, not prose.
- 3Open →Intermediate4–6 hours
Working With Your Own Documents and Data
The model does not know your policies, contracts, notes or database. Learn the three ways to give it that knowledge — paste it, upload it, or retrieve it — and when each one stops working.
No-code: Build a personal expert on a 200-page document with NotebookLM or a Project
Code: Minimal RAG: embed chunks, retrieve top-3 by cosine similarity, answer with citations. In-memory for clarity.
- 4Open →Intermediate4–6 hours
Automating Real Work: Tools, Actions and Custom Assistants
A model that can only write is a consultant; a model that can act — search, calculate, send, update — is a colleague. Learn how tools work, how to build a custom assistant without code, and how developers wire actions in.
No-code: Build a shareable "Customer reply assistant" for your team, no code
Code: Function calling end-to-end with one tool (order lookup). The loop is the thing to learn; the field names are trivia.
- 5Open →Advanced1–2 weeks
From Assistants to Agents: Multi-Step Work You Can Delegate
An agent is a model in a loop with tools and a goal. Learn to delegate multi-step work safely as a user, and to build the loop — then graduate to the full Agent Workflows track.
No-code: Delegate a research task, then a browser task — and check the work
Code: A minimal agent with two tools and an approval gate on the one that has side effects.
- 6Open →AdvancedOngoing
Trusting It: Evaluation, Governance and Going Deeper
Anyone can get a demo working; the advanced skill is knowing how good it is and keeping it that way. Learn to measure quality, manage risk, and choose what to learn next — fine-tuning, serving, or the theory.
No-code: Build a 25-case scorecard for your assistant and run it monthly
Code: A 40-line eval harness: deterministic checks first, LLM-as-judge for the rest, pass rate at the end.
A six-week plan (~5 hours a week)
One stage per week. Pick your lane; both columns end in the same place.
| Week | Non-developers | Developers |
|---|---|---|
| Week 1 | Stage 1. Watch Karpathy’s intro; do the 20-minute strengths/limits exercise; start the usage log. | Stage 1. Make the first API call with two providers; print token counts; write the summariser script. |
| Week 2 | Stage 2. Write and refine one four-part prompt for a weekly task; save it as a template. | Stage 2. Structured output with a schema; add few-shot examples; compare three model tiers. |
| Week 3 | Stage 3. Build a NotebookLM / Project on real documents; test ten questions with citations. | Stage 3. Minimal RAG on a folder of documents; measure retrieval recall separately. |
| Week 4 | Stage 4. Build and share a custom assistant with knowledge files and a draft-only rule. | Stage 4. Function calling with two tools and an approval gate; wrap them as an MCP server. |
| Week 5 | Stage 5. Delegate one task a day to a research, browser or coding agent; log the results. | Stage 5. First agent with ADK / Agent SDK / Agents SDK; read every trace. |
| Week 6 | Stage 6. 25-case scorecard for your assistant; one-page governance note. | Stage 6. Eval harness in CI; then start the Agent Workflows track. |
Finished the path?
Developers continue to the Agent Workflows track. Everyone else: you now know more about using AI well than most of the people deploying it — go build the scorecard for something real.