GenAI Engineering Roadmap

The production-grade, evals-first, career-aware roadmap for GenAI engineers in 2026.
From "I can call the OpenAI API" to "I shipped a multi-agent system that runs in production with SLOs and an evals harness."

---

What this is

A rigorously-sourced, file-per-topic, evals-first roadmap for becoming a production GenAI engineer. Every factual claim is cited and dated in LAST_VERIFIED.md. Every topic follows a consistent template. Every project ships with code, tests, and an evals harness.

This is not a link dump. This is not a single 600-line README. This is not LangChain-only. This is the roadmap a principal AI engineer would recommend without caveats.

What this is not

• Not a tutorial. Topics assume you can read Python and understand HTTP.
• Not exhaustive. We picked one lane — production GenAI engineering — and go deep.
• Not a model leaderboard. We cover architectures and patterns that survive model releases.

The 12-rung skill ladder

Rung	Topic	Folder	Est. time
0	Foundations	foundations/	1 week
1	Applied LLM Calling	llm-calling/	1 week
2	Context Engineering	context-engineering/	2 weeks
3	Modern RAG	rag/	3 weeks
4	Tools & MCP	tools-and-mcp/	2 weeks
5	Single-Agent Engineering	agents/	2 weeks
6	Multi-Agent Orchestration	multi-agent/	3 weeks
7	Evals & Test-Driven LLM Dev	evals/	2 weeks
8	Production & Economics	production/	2 weeks
9	Fine-tuning & Self-Hosting	fine-tuning-and-self-hosting/	2 weeks
10	Safety, Alignment & Red-Teaming	safety/	1 week
11	Real-Time & Voice	realtime-voice/	1 week

Pick your starting point

Your background	Recommended path
Backend engineer	docs/learning-paths/from-backend-engineer.md
Frontend engineer	docs/learning-paths/from-frontend-engineer.md
Data engineer	docs/learning-paths/from-data-engineer.md
Data scientist	docs/learning-paths/from-data-scientist.md
ML engineer	docs/learning-paths/from-ml-engineer.md
Starting from zero	docs/learning-paths/from-zero.md

The GenAI Field Guide

Career and industry context — separate from the technical curriculum.

• field-guide/01-the-ai-engineer-role.md
• field-guide/02-skills-that-get-you-hired.md
• field-guide/03-interview-loop-anatomy.md
• field-guide/04-portfolio-strategy.md
• field-guide/05-salary-bands.md
• field-guide/06-job-market-trends.md

Portfolio projects

15 portfolio-grade builds, each with README.md, src/, tests/, evals/, DEPLOYMENT.md. See projects/.

Decision guides

When you hit a fork in the road, start here:

• docs/decision-guides/choosing-an-agent-sdk.md
• docs/decision-guides/choosing-a-vector-db.md
• docs/decision-guides/rag-vs-long-context.md
• docs/decision-guides/fine-tune-or-not.md
• docs/decision-guides/self-host-vs-api.md
• docs/decision-guides/open-weight-model-selection.md

Curated external resources

awesome/ — papers, books, courses, blogs, tools, communities. One-line reviews, not bare links.

How to use this repo

1. Read LAST_VERIFIED.md to see when every factual claim was last checked.
2. Pick a learning path based on your background.
3. Work the rungs in order. Each rung links to the next.
4. Build at least 3 projects as you go. Code beats reading.
5. Run the evals that ship with each project. Internalize the evals-first mindset.

Contributing

See CONTRIBUTING.md. The short version: every factual claim must be sourced and dated, every file must follow the topic template, every project must ship with tests.

License

MIT — see LICENSE.

Acknowledgements

• alexeygrigorev/ai-engineering-field-guide for the field-guide playbook and the "data, not filler" philosophy.
• Every practitioner who has published a postmortem, evals harness, or agent architecture in public.

Star history

If this roadmap helps you, star it. Stars keep the project maintained.

Connect With Me