Roadmap
AI engineer
Build real products on top of LLMs: prompting, RAG, agents, evaluation and shipping.
Who it's for
Developers who want to build products with LLMs, whatever their background in machine learning.
Where it leads
- AI engineer
- LLM engineer
- Applied AI developer
- Full-stack AI developer
By the end you can
- Explain what an LLM can and cannot do, and why
- Write prompts that hold up with real users
- Build a RAG system and measure its quality
- Build an agent with tools, and keep it safe
- Ship and monitor an LLM feature for cost and latency
- 1
How LLMs work
Tokens, attention and transformers, enough to reason about a model's behaviour.
- 2
Prompting that holds up
Clear instructions, examples, structured output and defending against prompt injection.
- DocsPrompt engineering overviewAnthropic
- DocsOpenAI CookbookOpenAI
- ArticleOWASP Top 10 for LLM applicationsOWASP
Do this: Write a prompt for a real task, collect 20 test inputs and improve it until it passes all of them.
- 3
Embeddings and retrieval
Give models your data: chunk, embed, retrieve and ground the answer in sources.
- 4
Agents and tool use
When to use a workflow, when an agent, and how to connect tools safely.
- 5
Evaluate before you ship
Golden sets and model-graded checks turn "it seems better" into numbers you can trust.
- ArticleYour AI product needs evalsHamel Husain
- ArticleChip Huyen's blogChip Huyen. Author of AI Engineering (O'Reilly).
Do this: Build a 50-case eval set for your app and run it on every prompt change.
- 6
Run it in production
Cost, latency, monitoring and keeping up with a field that moves every week.
Learning this with others is easier.
Ask questions when you're stuck, find people on the same path, and track your progress on DevLearn.