AI Engineer
Understand what LLMs actually do, call them from code, and grow from single prompts to tool use, MCP, retrieval and agents with memory — then learn the evaluation, cost and security discipline that production AI features need.
- 17 lessons
- 3 stages
- 2 branches
- 18 concept notes
- includes Python (optional)
- includes Git & Collaboration
- includes SQL & Databases (optional)
Your progress
0 / 16
required steps
+ 0/2 optional
- 01
LLM foundations
What a model is, and how to talk to one from code.
- Py
Step 01 · Roadmap · 10 lessons · optional
Python
Most AI tooling is Python-first, and the recommended route below uses it. Skip this if you already write Python comfortably.
0/10 - 02How LLMs workTokens, next-token prediction, context windows and model choice.
Step 03 · Pick one
Call an LLM API from code
Same concepts, your language. Pick the one you build in.
- 04Prompt engineeringClear instructions, examples, context and iteration.
- 05Structured outputGetting validated JSON out of a model, every time.
- Gt
Step 06 · Roadmap · 10 lessons
Git & Collaboration
Prompts, evals and agent code are code: version and review them like any other.
0/9
- 02
Building with LLMs
Give models tools and knowledge they were never trained on.
- 07Tool useFunction calling and the request → tool → result loop.
- 08Model Context ProtocolBuild and connect MCP servers, and secure what they expose.
- 09EmbeddingsTurning text into vectors and measuring similarity.
- 10Vector searchpgvector, ANN indexes and metadata filtering.
- Sq
Step 11 · Roadmap · 13 lessons · optional
SQL & Databases
pgvector lives in PostgreSQL. Solid SQL makes retrieval far easier.
0/13 - 12Retrieval-augmented generationChunking, retrieval, reranking and grounded answers.
- 03
Agents & production
Ship AI features you can measure, afford and trust.
- 13AgentsAgent loops, planning, and when not to build one.
- 14Context & memoryBudgeting the window, compaction, sub-agents and long-term memory.
- 15EvaluationsTest sets, graders and catching regressions.
- 16Cost & latencyPrompt caching, batching, model routing and streaming UX.
- 17Security for AI appsPrompt injection, data leakage and safe output handling.
Step 18 · Pick one
Serve models in production
Most teams start on hosted APIs; some need to run open models themselves.
- You made it.