Start here¶
Answer these questions honestly. Start at the first module where you answer "no".
| # | Can you... | If no, start at |
|---|---|---|
| 1 | Explain what a token is, why LLMs hallucinate, and what RAG does? | 00 · Foundations |
| 2 | Call an LLM API from Python and read the token usage? | Calling an LLM from code |
| 3 | Use a coding agent with a plan, tests first, and review, and write an AGENTS.md? |
01 · Coding agents |
| 4 | Build an agent that calls tools, with a step limit and a human approval step? | 02 · Building agents |
| 5 | Explain indirect prompt injection and why a better prompt doesn't fix it? | 04 · LLM security |
| 6 | Design input, output and action guardrails, and mask PII before logging? | 03 · AI safety and guardrails |
| 7 | Build a tagged eval set and an LLM-as-judge you've checked against your own labels? | 05 · Evaluation |
| 8 | Design agent memory that can't be poisoned by a web page? | 06 · Agent memory |
| 9 | Trace an agent, report p95 latency and cost per task, and load test it? | 07 · Agent ops |
| 10 | Write a design doc and threat model for an AI feature, including DPDP and EU AI Act impact? | 08 · System design |
Answered "yes" to everything? Go to 09 · Careers, build the projects your own way, and consider contributing.
What you need¶
- Time: about 8 to 10 hours a week for 90 days covers everything. See the 90-day plan.
- Skills: basic Python (functions, lists, dictionaries, installing packages). You don't need machine learning or maths beyond school level.
- Tools: a laptop with Python 3.12+, uv, and Git. Docker is optional (only for P5).
- Money: none. Every project's tests run offline. A free Gemini API key lets you try real models, and an entry-level coding-agent plan (around $20 a month, or a free tier) is enough for module 01.