Skip to content

AI Engineer roadmap: mind map

A mind map of the full roadmap.sh/ai-engineer topic set (169 nodes, 17 branches), for planning and scheduling. Three formats, use whichever your tool likes:

File Open it with
ai-engineer-roadmap.xmind XMind: double-click to open directly.
ai-engineer-roadmap.opml XMind, MindNode, Freeplane, OmniOutliner, most mind-map tools (File → Import → OPML).
ai-engineer-roadmap.md Any editor (readable as-is); XMind also imports Markdown (File → Import → Markdown).

If the .xmind ever won't open in your XMind version, import the .opml or .md instead, same content. To edit the map, change ai-engineer-roadmap.md and re-generate, or just edit in XMind.

Full node-by-node audit: see coverage.md: every one of the 169 nodes mapped to where it's taught.

Roadmap section → course module (for scheduling)

Where each roadmap branch is taught in this course (covered · partial · planned · out of scope):

Roadmap branch Course module
Introduction (what AI eng is, vs ML eng, AGI) M0
How LLMs work (tokens, context, inference) M0 (builder's depth; not transformer internals)
Models & selection (closed/open, families) M0 + M4/M6
Using model APIs (messages, structured outputs) M4, M6
Prompt engineering (few-shot, CoT) M5
AI safety & ethics (bias, moderation, adversarial) M10 (security) + planned Ethics module
Open-source & local models (HF, Ollama) planned (M3 PyTorch box only today)
Embeddings M7
Vector databases (Chroma, FAISS…) M7 (Chroma; others surveyed)
RAG (chunking, retrieval, re-ranking) M7-M8
AI agents (function calling, frameworks) M9 (LangGraph deep; rest surveyed)
MCP (client/server) M9 (surveyed; not built)
Multimodal AI (image/audio/vision) planned (this run)
Classic NLP tasks woven into M5/M6 examples
Evaluation, optimization & monitoring M8 (RAG eval), M11 (monitoring)
Fine-tuning & training out of scope by design (we use hosted models)
Data & integrations (SQL, web search) light / out of scope

This course is deliberately the applied, beginner path: it doesn't mirror the roadmap node-for-node (that's a reference map, not a curriculum), but it covers the core build skills and points to the roadmap for breadth.