aiengineering.guideaiengineering.guide

The Path

Where to start, and what depends on what

Lessons are grouped into tracks and ordered by prerequisite. You don't have to go in order — every lesson is standalone — but if you want a route, here's the map.

Lesson prerequisite graphLessons are arranged in columns by track and rows by order; arrows point from a prerequisite to the lesson that depends on it.0 · Foundations2 · Talking to Models3 · Tools, Agents & MCP4 · Retrieval & Long Context5 · Production (MLOps + LLMOps)7 · Career transitionsWhat is an LLM, really?Prompting, as a real sk…The Python you actually…Reading a model card an…Your first real API callStreaming responses, ho…Structured output you c…Tokens, context, and wh…Planning strategies tha…Tool use, from three li…Long-running agents and…The agent loop, from sc…MCP, in the shape you'l…When agents should give…Embeddings, without the…Chunking is the decisionVector stores worth usi…Evaluating a RAG system…Offline evals that catc…Prompt injection is a r…Observability that surv…Caching and cost controlDesign an LLM gatewayThe three portfolio pro…What AI engineering int…Reading papers without …

Where to start

Zero prior AI experience
Read straight through Stage 0 → 6. Skip nothing on the first pass. If a maths or Python reference loses you, hop to /foundations for a pointer, then come back — the roadmap assumes only the shape of the reference, not fluency in it.
Working software engineer
Skim Stage 0 (mostly familiar) and Stage 1 (mostly familiar API mechanics with LLM quirks). Start reading in earnest at Stage 2. Stage 5 (Production Reality) is likely where the roadmap adds the most new material.
Data scientist / ML practitioner
You know the models. The gap is shipping. Skip to Stage 2 (tools) and Stage 5 (production) — evals, guardrails, observability, cost — then loop back to Stage 3 (RAG) if retrieval is new to you.

Missing the fundamentals a lesson assumes? Start with Foundations →

Enter to go · Esc to close