New: Roadmaps ordered paths through our cheat sheets and flashcards, so you always know what to study next.
Explore themSee what's new on GitHubFrom scoping a client's AI problem to running it as their production system.
A 13-step learning path. Follow it in order, or jump to what you need.
This path is for AI engineers ready to take their skills out of the notebook and into a paying customer's own environment: scoping the real problem, proving the system works, and keeping it healthy after launch. Expect about 6 to 10 weeks at a few hours a week, moving from stakeholder scoping and business-case framing through AI governance, production observability, and evaluation engineering, into the API security and infrastructure choices you inherit once you deploy inside someone else's stack. It does not re-teach prompting, RAG, or agent building, which stays the territory of AI Engineer (LLM Applications); this path picks up exactly where that one ends. By the end you can scope an AI engagement with a client's stakeholders, prove a production system is working with real evaluation evidence, monitor it once it's live, and defend its ROI and compliance posture to people who never touch code.
Expected: the core LLM and RAG engineering skills from AI Engineer (LLM Applications), prompting, retrieval, agents, and evaluation basics. Helpful but not required: previous client-facing or consulting experience.
Start here: sharpen how you listen and explain before you touch a line of client code, the same instinct every scoping conversation in this path leans on.
Take the listening habits from step 1 and turn them into a repeatable way to find out who actually holds power over your engagement, before you waste a sprint building for the wrong person.
Turn the people you just mapped in step 2 into a real specification: elicit, prioritize, and trace requirements so what you build matches what the client actually asked for.
You can read a room, map who actually holds the decision, and turn a vague ask into requirements you can build against. Next up: making the case for why any of it is worth building.
Finish this section to unlock.
+100 XP
Before you propose anything, translate the system from step 3 into a client's own numbers: revenue, cost, or risk, so an executive says yes instead of asking what it's for.
Any system you're now justifying financially also has to survive a compliance review: map the EU AI Act tier and data-protection obligations a regulated client will raise before you write a proposal, not after.
This is where production AI stops behaving like normal software: traces, spans, and quality checks replace the pass or fail tests you're used to, and most engineers have to revisit this mental model more than once before it settles.
You can put a dollar figure on the system you're proposing, name the compliance guardrails a regulated client will ask about, and trace a live request the way production AI actually needs watching. Next up: proving the thing is actually good, not just running.
Finish this section to unlock.
+100 XP
The traces from step 6 tell you something happened; this turns that into proof it happened well, building the release-gate evidence a skeptical client engineering team will actually accept.
A system you can prove works still has to plug into a client's existing stack without opening a hole in it: close the authorization and injection gaps that make APIs the easiest way into someone else's data.
Take this if your next engagement hands you a client's own servers to deploy into: package the system from earlier steps into a container so it behaves identically on their infrastructure as it did on yours.
Go further if that client infrastructure already runs at cluster scale: orchestrate the container from step 9 across machines instead of babysitting it on one.
A few minutes of due flashcards on chapter 1 keep that scoping work fresh while you build on it: you can now score a system against real evaluation evidence instead of a gut feeling, and close the authorization gaps that turn an API into a breach. Next up: picking up whatever infrastructure your next client already runs.
Finish this section to unlock.
+100 XP
Pick this up if the client you land on runs AWS, the most common cloud a forward deployed engineer inherits: the same patterns transfer to Azure or GCP if that's what you find instead.
Reach for this if your delivery leans heavily on multi-agent systems: extend the observability habits from step 6 to the coordination, handoffs, and drift that only show up once several agents run together.
Pull governance, observability, evaluation, and security into one production posture: the same end-to-end ownership that separates a forward deployed engineer from everyone who hands off before launch.
You can scope a client's problem, prove the system works, watch it in production, and defend its cost and compliance to the people signing the contract, the same end-to-end ownership the role is built on. This badge marks that you can own an AI system inside someone else's walls, not just build one in your own.
Finish this section to unlock.
+100 XP
Finish every required step, at least 70% of them genuinely done (not skipped), to earn this badge and 500 XP.