An AI Academy for the People Who Keep the Lights On

Every AI course is written for someone else

If you work in infrastructure support at a global systems integrator — TCS, HCL, Infosys, Wipro, Cognizant, Capgemini — you have almost certainly been told, more than once this year, to "upskill in AI." You have probably also opened one of the recommended courses, watched forty minutes of someone training a sentiment classifier on a movie-review dataset in a Jupyter notebook, and quietly closed the tab.

Not because it was bad. Because it was for someone else.

There is a large, capable, and strangely underserved population of technologists: the people who run Intune and SCCM estates, work ServiceNow queues, patch Windows and Linux fleets, manage M365 tenants, run transitions and migrations, and hold P1 bridges together at 2am. They keep the lights on for the Global 2000. And when they go looking for AI education, they are offered one of two things — a data science curriculum that assumes Python and linear algebra, or a corporate LMS module that is mostly a compliance video with a quiz.

That gap is not a small one. Infrastructure support is exactly where agentic AI is landing hardest and fastest — ticket deflection, automated triage, self-healing endpoints, AI-drafted knowledge. The people closest to that change have the least education designed for them. Being closest to the disruption and last in line for the training is a bad place to stand.

So I built the thing I couldn't find.

Engineer + AI = better engineer

The whole platform rests on one conviction, and it's worth stating plainly because it shapes every design decision underneath it: AI is not coming to replace the infrastructure engineer. It is coming to rewrite what the infrastructure engineer does all day. The people who understand that early — and can prove it with artifacts — are the ones who move up. The people who wait to be told are the ones who get restructured around.

Three principles follow from that.

Contextual, not generic

A course for service-desk engineers should not explain "hallucination" with a poem-writing example. It should explain it the way it will actually bite you: an LLM is one more confident witness, exactly like the user who swears they "didn't change anything." You already have a professional instinct for confident-but-wrong information. The course connects the new concept to the instinct you already own.

Applied, not watched

Every module ends with something you did, not something you saw. There is a project, a set of drills, and an Apply-at-Work mission — a task performed in your actual job that week, on real (sanitised) tickets, real logs, a real status report. Learning that never leaves the browser doesn't survive contact with Monday.

Honest about limits

Roughly a third of the content is about when not to use AI: anonymisation before any ticket text is pasted, validating generated scripts before they touch production, knowing which decisions must stay human. An AI course for people with production access has an obligation to teach restraint, not just capability. There is a downloadable template in the Service Desk course literally called "When NOT to Use AI" Checklist.

You don't enrol in a course. You start a job.

This is the piece I'm proudest of, and the one that most clearly separates AICademy from everything else on this list.

When you open the Service Desk programme, you are not shown a syllabus. You are shown your new employer: Northwind Financial, a 10,000-employee financial-services firm with offices in Sydney, Singapore and London. You're on the Service Desk. The queue never sleeps.

You get the stack you'll be supporting for the next eight weeks — Windows 11, Intune + SCCM co-managed, Entra ID, M365, ServiceNow, GlobalProtect VPN, CrowdStrike, Splunk, Nexthink. And you get colleagues, with names and personalities:

Four of the six courses have a world like this, each matched to its discipline — Northwind Financial for the service desk, Meridian Logistics for the sysadmin, Northgate Retail Group for the project manager, Alderbrook Health for the model-internals course. The characters recur. The environment persists. By week six you know that estate the way you know a real account.

Every single module then opens with Today's Queue from that environment. Not an abstract exercise. A shift:

That lens callout changes name by course, because the reframing is the whole point. The same underlying concept arrives as a Frontline lens for the service desk, an Ops lens for the sysadmin, a Delivery lens for the PM, an Infra lens for the RAG engineer, a Leadership lens for executives, and a Decision lens for the model-internals course. Six audiences, six vocabularies, one idea.

Six programmes, 82 modules

The courses are deliberately pitched at different rungs of the same ladder. You join where your job is now — nobody has to start at the bottom.

Two more are queued: Copilot & M365 AI for the Workplace and AI Security & Governance for Enterprises.

Six steps, every single week

Every module in every course runs the same loop. The consistency is deliberate: once you learn the rhythm in week one, you never spend attention on navigation again — all of it goes to the content.

And the things you keep

Beyond the weekly loop, each course carries a toolkit of downloadable markdown templates — the artifacts that outlive the course and go straight into your working life or your portfolio.

Deliberately small

The architecture is boring on purpose. One repo, one deployment, one website; every course is a folder. Adding a course means creating a data directory and registering it in one file — the homepage card, routes, progress tracking, XP, certificates, tutor and flashcards all light up automatically.

A few choices worth calling out, because they were arguments I had with myself:

Content as code, not a CMS. Every course is plain JavaScript data files in the repo. That means the curriculum is versioned, diffable, reviewable in a pull request, and impossible to lose to a database migration. It also means writing a course feels like engineering, which — for this material — is the correct feeling.

Retrieval that runs in your browser. The in-app assistant does TF-IDF lexical retrieval over the course corpus — modules, lenses, resources, rubrics, quiz explanations, missions and published notes — entirely client-side. No vector database, no per-query cost, no data leaving the machine. It is, fittingly, the "naive RAG" that the flagship course teaches you to build in week six.

Offline by default. Supabase handles auth and cloud sync when it's configured; when it isn't, everything falls back to localStorage and the whole platform still works. A PWA manifest and service worker mean a learner on a bad connection — or on a locked-down corporate laptop — still gets the course.

Against the courses you've already tried

I want to be fair here: the big platforms are excellent at what they do. Coursera and DeepLearning.AI produce world-class material. The difference isn't quality — it's who the material assumes you are, and what it expects you to do with it.

The rank ladder in the flagship course makes the intent explicit. It runs L1 Service Desk → Desktop Support Engineer → Sysadmin → Endpoint Engineer → Platform Engineer → Hybrid Cloud Architect → Automation Lead → AI Ops Engineer → Agentic AI Engineer. That's not a gamification gimmick. That's the actual career staircase in a GSI, with the last two rungs added — the two that didn't exist five years ago and now command the premium.

Eight weeks later, what's actually different about you?

This is the question I held every design decision against. Not "what will they have watched" — what will have changed. Six answers:

If you want to look around before then, the whole thing is live and free — no signup required to browse the courses.

▸ Explore AICademy Read the full course sources

Built by Ajay Walia. If you work in infrastructure at a GSI and something here is wrong about your day-to-day, I want to hear it — that feedback is what the next course gets built from.