Become an AI engineer by building.

Structured tracks from beginner to AI engineer. Coding labs that run in your browser. An AI mentor that sees your code and guides you to the fix. A verified Passport that proves what you can do.

Sign in with GitHub. No setup, no credit card.

Lab · Build a vector index from scratch Run

Requirements

  • cosine(a, b)
  • normalize(v)
  • VectorIndex.add(id, vector, metadata) stores items; adding an existing id replaces it
  • VectorIndex.search(query, k, where=None)
def cosine(a, b):
    dot = sum(x * y for x, y in zip(a, b))
    na = math.sqrt(sum(x * x for x in a))
    nb = math.sqrt(sum(y * y for y in b))
    return dot / (na * nb)  # ✗ zero vector?

AI mentor: Your test cosine handles zero vector fails with ZeroDivisionError. What is the length of [0, 0] — and what should similarity mean when a vector has no direction?

4career tracks
52modules
37graded coding labs
6portfolio projects
300interview questions

Try a real lab — no sign-in

Write your first Python. Run the tests.

A real beginner lab, exactly as learners see it. Fill in the functions and press Run.

Python from zeroLab · Your first gradebook

Make these pass

  • average(scores) returns the mean of a list of numbers, rounded to 1 decimal place; an empty list returns 0.0
  • letter_grade(score) returns A for 90+, B for 75+, C for 60+, D for 40+ and F below 40
  • top_student(marks) takes a dict of name → score and returns the name with the highest score (alphabetically first on a tie)
  • passed(marks, cutoff=40) returns the names that scored at least cutoff, sorted alphabetically
  • count_words(text) returns a dict of lowercase word → how many times it appears, ignoring . , ! ?

Runs in your browser. In the Academy, hidden tests grade it in a sandbox and the AI mentor helps when you're stuck.

Save your progress

An AI mentor for every learner

It doesn't hand you the answer. It makes you the engineer who finds it.

The mentor sees your course, your previous attempts, your code, your quiz results and your project — and switches roles as you need.

01

Teacher

Explains the concept behind your bug with a tiny example — not your answer.

02

Debugger

Reads your code and failing tests, asks diagnostic questions, points to the line.

03

Reviewer

Reviews your repo like a senior engineer: correctness, evals, security.

04

Interviewer

Runs a mock interview on the topic you just learned, then scores you.

No setup. Just build.

AI engineering has a huge environment problem. Labs run Python right in your browser, and graded submissions run in an isolated sandbox.

Python
Docker
GPUs
Vector DB
API keys
Environments

4 tracks · one clear path each

See every module before you start.

Pick one track. Its modules unlock in order, so you always know the next step — and projects appear exactly when you're ready to build them.

AI for University Students

From your first Python script to a portfolio of shipped AI projects.

Best for: College students and fresh graduates in any branch

23

modules

15

labs

3

projects

  • Write clean, tested Python and ship it through GitHub
  • Explain how LLMs, embeddings and transformers actually work
  • Build and evaluate a RAG application on your own documents
  • Ship a tool-using agent and a portfolio that proves it

AI Engineer Intern · Junior AI Engineer · GenAI Developer · AI Solutions Associate

Modules, in order ~76h of guided work
  1. 01What AI is, in plain wordsA friendly map of today's AI — what the common words mean, what kinds of AI exist, how AI products are put together and who builds them. No code, no maths.40m
  2. 02Python from zeroYour first steps in Python — values, lists, dictionaries, decisions, loops and functions — using everyday examples like marks and word counts.90m
  3. 03Clean, reliable PythonThe next step after the basics — type hints, dataclasses, generators, decorators and clean error handling, the features every serious Python codebase uses.75m
  4. 04Git and GitHub for AI projectsBranches, pull requests, clean history and a README that makes a reviewer — or a hiring manager — trust your project.45m
  5. 05Working with messy dataParse, clean, normalise and de-duplicate the kind of real-world data AI systems are actually fed — and ship it as JSONL.60m
  6. 06Vectors, similarity and probabilityThe handful of maths ideas behind modern AI — dot products, vector length, softmax and cross-entropy — built by hand with small examples.60m
  7. 07Machine learning fundamentalsTraining, testing, overfitting and the metrics that matter — built from scratch so you can reason about any model.75m
  8. 08Neural networks and transformers, intuitivelyFrom a single neuron to self-attention — what a transformer actually computes, and why that explains LLM behaviour you see every day.75m
  9. 09How LLMs work: tokens, context and samplingWhat actually happens between your prompt and the answer — tokenisation, context windows, prefill and decode, temperature, top-k and top-p.60m
  10. 10Working effectively with AI assistantsUse chat assistants and coding agents like a professional — clear briefs, context, verification, and knowing what not to delegate.40m
  11. 11APIs, HTTP and JSON contractsHow every model call really works — requests, status codes, retries, streaming with Server-Sent Events — so you can build on any provider.60m
  12. 12Prompt engineering and structured outputWrite prompts like an engineer — templates, few-shot examples, JSON output you can validate, and repair loops when the model gets it wrong.75m
  13. 13Calling LLMs reliably: retries, fallbacks and costProduction LLM calls fail, stall and cost money. Build the retry, fallback and budget machinery that keeps an AI feature up and affordable.70m
  14. 14Project: a production-style AI API ProjectBuild a model-comparison service with validated structured output, retries, fallbacks, cost tracking, tests and a container — your first portfolio-grade AI backend.900m
  15. 15Embeddings and vector searchHow text becomes vectors, how similarity search really works, and where the abstraction leaks.60m
  16. 16Ingestion and chunkingTurn raw documents into retrievable chunks — section-aware splitting, overlap, contextual headers, stable IDs and de-duplication.65m
  17. 17Grounded generation: context, citations and saying 'I don't know'Turn retrieved chunks into trustworthy answers — context budgets, grounded prompts, verifiable citations and principled abstention.70m
  18. 18Project: an evaluated RAG application ProjectBuild a RAG assistant over real documents with hybrid retrieval, citations, abstention and a measured evaluation — the most-requested AI engineering skill.1200m
  19. 19Tool and function callingLet models call your code safely — JSON Schema tool definitions, argument validation and a dispatcher that never crashes the conversation.70m
  20. 20Agent fundamentals: the loopWhat an agent really is — a model in a loop with tools and a stopping rule — and when a simple workflow is the better choice.70m
  21. 21Responsible AI in practiceFairness, transparency, privacy, safety and accountability — turned into concrete engineering habits you can apply on every project.45m
  22. 22Your AI portfolio and job searchTurn what you build into interviews — a portfolio of verified projects, a sharp resume, a public Passport and a preparation plan for AI engineering interviews.50m
  23. 23Project: a tool-using agent ProjectBuild an agent that completes a real multi-step task with tools (including an MCP server), guardrails, human approval and trajectory evaluation.1200m

Inside a lesson

Plain English first. Then code.

Every module is a short, focused lesson written for builders, followed by a knowledge check, a graded lab and interview questions on the same topic.

  • Store and change values using numbers, strings, lists and dictionaries
  • Make decisions with if, elif and else
  • Repeat work with for loops
  • Write your own functions and read Python error messages without panic

Python from zero · 90 min

Why start here

Every AI system you will build later in this track is, underneath, an ordinary Python program: it reads some data, makes decisions, repeats steps and returns a result. If you are comfortable with the ideas in this module, everything after it is about what to build, not about fighting the language.

No setup is needed. The lab at the end runs Python right in your browser.

Values and variables

A variable is a name for a value:

name = "Asha"        # a string (text) — always in quotes
marks = 82           # an integer (whole number)
attendance = 0.93    # a float (decimal number)
is_topper = True     # a boolean: True or False

You can change a variable at any time, and combine values:

marks = marks + 5            # now 87
greeting = "Hello, " + name  # "Hello, Asha"
print(greeting)

print(...) shows a value on the screen. It is the simplest way to see what your program is doing.

Lists: many values in order

A list holds several values in order, inside square brackets:

scores = [82, 67, 91, 45]

scores[0]          # 82  — positions start at 0
scores[-1]         # 45  — negative positions count from the end
len(scores)        # 4
scores.append(73)  # add to the end
sum(scores)        # 358

Portfolio projects

Build things you can show an interviewer.

Each project lives in your own GitHub repo. An AI reviewer checks it against the acceptance criteria at a pinned commit — pass, and it's verified on your Passport.

Project: a production-style AI API

A small web service that sends the same task to several AI models, checks the answers are in the right format, and records how good, fast and expensive each one was.

  • docker compose up starts everything with only an .env file
  • Invalid LLM JSON is caught and retried; failure returns a clear 4xx/5xx
  • Benchmark table covers 50 examples x 3 models with accuracy, p50/p95 latency and cost

University

Project: an evaluated RAG application

An assistant that answers questions from a set of documents, shows exactly where each answer came from, and has a scorecard proving how often it is right.

  • Golden set of 100 questions including 15 unanswerable ones
  • Experiment table with at least 4 configurations and recall@5, MRR, faithfulness, latency, cost
  • Every answer shows citations that open the right source

University · FDE · Generalist

Project: a tool-using agent

An AI assistant that can take actions using real tools (search, database, calendar, tickets), asks for permission before risky steps, and can be tested repeatedly to see how reliable it is.

  • Design doc explains why workflow steps vs agent steps
  • MCP server usable from at least one other MCP client
  • Benchmark of 20+ tasks run 5 times each with pass^1 and pass^5

University · FDE · Generalist

Capstone: a serious domain-specific AI system

One large project that looks like something a real company would build and pay for, solving a clear problem for a clear group of users, with everything you learned combined.

  • Passes the capstone rubric at 80% or higher
  • Real or realistic data, not toy examples
  • Deployed demo, 3-minute video and full documentation

FDE

Capstone: AI architecture dossier

A complete architecture package for an enterprise AI system — the documents an architecture review board expects, plus a small working slice that proves the riskiest decision.

  • Requirements state measurable quality, latency, scale, cost and compliance targets
  • The architecture shows data flows, trust boundaries, identity propagation and where policy is enforced
  • Estimates are computed with stated assumptions and checked against the latency budget and cost targets

Architect

Project: a multimodal system

An AI system that works with more than text: it reads scanned documents, understands images, or talks with users by voice.

  • A: field-level precision/recall per field on 50+ documents; low-confidence routing to human review
  • C: end-to-end response latency p50 and p95 with per-stage breakdown; barge-in works
  • Cost per document or per minute of conversation reported

Generalist

Learn → build → prove

Not a certificate. A Passport of evidence.

Sandbox-graded labs

Every lab submission runs with hidden tests in an isolated microVM. No copy-pasted screenshots.

Verified projects

Projects live in your own GitHub repos. AI code review checks them against real acceptance criteria at a pinned commit.

AI Anytime Passport

A public profile where every skill links to its evidence — plus certificates anyone can verify.

How a skill gets verified

Every skill on a Passport shows how far it has been proven — and links to the evidence.

  • Checked — passed the module's knowledge check
  • Lab-verified — passed a sandbox-graded lab with hidden tests
  • Project-verified — used in a reviewed project the learner owns

Top learners earn community credits.

Weekly challenges and a monthly leaderboard. Each month the top learners earn AI Anytime community credits to spend on perks.

Catch up with Sonu Present on the AI Anytime YouTube channel Research participation Job referrals Internship opportunities

For hiring teams

Hire on what they've built.

Search learners who are open to work by verified skill, project scores, sandbox-graded labs and challenge results. Shortlist them and request an intro — the learner decides whether to share contact details.

  • Search by verified skill, track and project
  • See the evidence: repos, reviews, scores
  • Shortlist and request intros in one place
Start hiring
junior AI engineers who can build RAG systems
Vector, keyword and hybrid searchCitations and groundednessOpen to workProject-verified

Results rank learners by evidence: verified projects first, then sandbox-graded labs, then knowledge checks — with links to every repo and review.

Questions

Before you start.

Is it free?

Yes — sign in with GitHub and start any track. Lessons, checks, browser labs, projects and the Passport are included. Heavy AI usage has a daily limit; you can add your own model key in Settings to go beyond it.

Do I need to install anything?

No. Labs run Python in your browser and graded submissions run in an isolated cloud sandbox. For projects you'll use your own GitHub repo and whatever editor you like.

Which track should I pick?

Every track above says who it's best for and lists all its modules. Choose the one that matches the job you want next. Your dashboard then focuses on that track only; you can switch later in Settings.

Why do modules unlock in order?

Each module builds on the last, and your dashboard always shows exactly one next step. Projects are milestones — they open once you've finished the lessons before them.

What is XP?

XP measures verified work: reading a lesson, passing a check, passing a lab in the sandbox, getting a project verified. It drives the monthly leaderboard, and top learners earn community credits for perks.

How do hiring teams use it?

They search learners who are open to work by verified skills and projects, review the evidence, and request an intro. You decide whether to share your contact details.

Your first lab is ten minutes away.

Sign in with GitHub, pick a track, and write your first line of AI engineering code — in the browser.

Start learning