Free, self-paced, updated for 2026

Become an AI engineer in 40 weeks.

One route from Python to production agents. Every concept explained and animated, every week mapped to the best free videos, and a plan that re-dates itself around the hours you actually have.

Start week 1 See the roadmap

245 lessons · 22 simulations · 129 videos · 520 hours of planned work

12345678
  • Prerequisites
  • Machine Learning
  • Other Modalities
  • Deep Learning
  • Practical Deep Learning
  • Build a GPT
  • AI Engineer Track
  • Portfolio & Job Search

Each station is a week. Hover one to see what it teaches.

Everything a self-taught AI engineer needs, in one place.

Not a link dump. Each week is a complete unit: lessons written to teach, simulations to play with, the right videos in the right order, practice, interview prep, and a self-check. Then your progress, tracked across all of it.

  • Lessons with intuition, formulas, code, pitfalls and a going-deeper section
  • The best article, paper or video hand-picked for every single lesson
  • Projects that prove each phase: two flagships with evals and cost analysis
  • Dates, streaks, a calendar heatmap and an honest ahead-or-behind count
What's on boardAll lines · all stops

The route, phase by phase

Foundations first, because every evaluation instinct an AI engineer relies on is born in classical ML. Then deep learning up to the transformer, a GPT built by hand, and twelve weeks on the work itself: LLMs, RAG, agents, evals, LLMOps and safety.

Week 1Week 20Week 40
  1. 1Weeks 1–5 · 36 lessons

    Prerequisites

    Python, SQL, AI-assisted coding, maths and the data stack.

  2. 2Weeks 6–11 · 35 lessons

    Machine Learning

    Fundamentals, feature engineering, algorithms and the first deployed project.

  3. 3Weeks 12–14 · 17 lessons

    Other Modalities

    Classical NLP in full, computer vision trimmed toward multimodal, time series as awareness.

  4. 4Weeks 15–19 · 27 lessons

    Deep Learning

    Neural networks from the perceptron up to the transformer block.

  5. 5Weeks 20–22 · 17 lessons

    Practical Deep Learning

    PyTorch, Hugging Face, fine-tuning and deployment.

  6. 6Weeks 23–25 · 16 lessons

    Build a GPT

    micrograd, makemore, a GPT and a BPE tokenizer, all typed from scratch.

  7. 7Weeks 26–37 · 81 lessons

    AI Engineer Track

    LLMs, prompting, context, RAG, agents, evals, LLMOps and safety.

  8. 8Weeks 38–40 · 16 lessons

    Portfolio & Job Search

    Polish eight projects, reposition, drill interviews, apply.

Open the full roadmap with every week

Inside every week

Eight stops, in the order that makes things stick: understand it, see it, do it, explain it.

  1. Stop 1

    Goal and done-when

    What the week is for, why it matters, and the test that says you're ready to move on.

  2. Stop 2

    Lessons

    Every concept explained with intuition, formulas or code, pitfalls, a going-deeper section and the best links.

  3. Stop 3

    Simulations

    Drag, slide and step through the hard ideas: attention, backprop, gradient descent, RAG and more.

  4. Stop 4

    Day-by-day plan

    Seven dated tasks that follow your pace, with status, hours and notes for each day.

  5. Stop 5

    Videos

    The right playlists and lectures for that week, playable in place.

  6. Stop 6

    Practice

    Warm-up, core and stretch exercises that turn reading into skill.

  7. Stop 7

    Interview prep

    How this week's material gets asked, with outlines of strong answers.

  8. Stop 8

    Self-check quiz

    Five questions with explanations, scored and saved.

Ideas you can play with

The hardest concepts come with live simulations. This one is real: click a word, drag a slider, press play.

All 22 simulations

From week 19. Every simulation also appears inside its lesson.

Built on the best free teaching there is

The roadmap structure follows the CampusX AI roadmap (AI Engineer fork), re-sequenced into 40 weeks. The videos come from the people who teach this best, and each one is placed on the exact week it's needed.

Core playlists and lectures

Every creator on the route

  • StatQuest40
  • Andrej Karpathy14
  • 3Blue1Brown14
  • CampusX9
  • Corey Schafer7
  • DeepLearning.AI6
  • freeCodeCamp5
  • Stanford Online4
  • Anthropic4
  • Umar Jamil3
  • Hamel Husain & Shreya Shankar3
  • Fireship2
  • Yannic Kilcher2
  • Hugging Face2
  • LangChain2
  • Tech With Tim1
  • James Briggs1
  • CodeEmporium1
  • Alfredo Canziani1
  • Carnegie Mellon University1
  • fast.ai1
  • Sebastian Raschka1
  • AI Engineer conference1
  • Edward Hu1
  • DeepFindr1
  • Chris Alexiuk1
  • Simon Willison1
Browse all 129 videos by week

Books and courses it leans on

Plus 129 docs and tools, 5 landmark papers, and 533 hand-picked links inside the lessons. Every video and link is checked automatically.

Finish with a portfolio, not a certificate

Eight projects prove each phase. The two flagships ship with eval results and a cost analysis, the two things almost no applicant shows.

  1. Project 1 · week 11

    Tabular ML pipeline, deployed

    Take a messy real tabular dataset from raw CSV to a prediction served over HTTP, with no manual steps in between.

  2. Project 2 · week 13

    TF-IDF text classifier (the LLM baseline)

    A TF-IDF + logistic regression classifier with recorded metrics. This becomes your permanent baseline: every later LLM approach gets compared against it.

  3. Project 3 · week 14

    Document understanding via a vision model

    Turn a document image (invoice, receipt or form) into validated structured JSON.

  4. Project 4 · week 28

    Schema-validated extraction CLI

    Document in, validated Pydantic JSON out, built the way you would build a production service.

  5. Flagship · week 32

    Flagship 1: production RAG with evals

    A RAG system that cites its sources, says 'I don't know' when it should, and comes with numbers that prove it works.

  6. Project 5 · week 33

    QLoRA fine-tune with an honest eval

    One QLoRA run on a small open model, with a before/after evaluation you would defend in an interview.

  7. Flagship · week 35

    Flagship 2: agent + MCP server

    A multi-step agent that is safe to leave running: tools, memory, approval gates and a hard cost ceiling.

  8. Optional · week 37

    Optional: Spring AI RAG service (Java bridge)

    Rebuild the core of Flagship 1 in Spring AI so enterprise JVM teams can read your résumé at a glance.

Your pace, not ours

Life happens. Tell the platform how many hours you have, and every date re-plans: the calendar, the week pages, your daily task. Fall behind and it tells you, honestly, what to cut and what never to cut.

  • Today view with your next task and ahead/behind
  • Calendar heatmap of every study day
  • Notes and actual hours on every day
  • Export and import your progress

Starting Mon, 12 Oct 2026, you'd finish around Sun, 18 Jul 2027, about 40 weeks.

The pace engine keeps every task in order and re-dates all 280 days around your hours. Change it any time in Settings; the calendar, week pages and dashboard follow.

Free to start. One small payment to finish.

Videos are never behind a paywall: they belong to their creators. Pro pays for what we write and build, once.

Free, forever

  • The full roadmap, calendar and pace engine
  • All 129 videos and 129 resources
  • Weeks 1–5 complete, with practice and quizzes
  • Lesson titles and plans for every week

Pro ₹499or $9 · once

  • Every lesson in weeks 6–40, with going-deeper sections
  • All 22 simulations
  • Practice, interview prep and quizzes for all 40 weeks
  • Sync across devices, plus every future update
See pricing

Questions

Who is this for?

Developers who want to become AI engineers: people who build LLM-powered products, RAG systems and agents, and run them in production. It assumes you can program. Engineers coming from backend work get extra tips throughout, because production skills are exactly what most AI teams lack.

Where does the roadmap come from?

The structure follows the CampusX AI roadmap's AI Engineer fork, re-sequenced into a 40-week plan. It adds a 'Build a GPT from scratch' phase (Karpathy's Zero to Hero), trims computer vision toward multimodal models, and puts extra weight on evals and LLMOps. Every lesson is written for this platform.

Which videos does it use?

Free YouTube material from the best teachers in the field: CampusX's 100 Days of ML and Deep Learning, 3Blue1Brown, StatQuest, Andrej Karpathy, Stanford (CS224N, CS25, CS336, CS229), DeepLearning.AI, Hugging Face, LangChain, Anthropic and more. Each video sits on the week where it's needed, and every link is checked automatically.

How much time does it take?

The default plan is 13 hours a week (two hours on weekdays, three on Saturday, Sundays off) for 40 weeks, about 520 hours. Set your own hours and the plan re-dates itself: 8 hours a week takes about 65 weeks, 20 hours about 27.

Does it cost anything?

The core is free forever: the whole roadmap, every video and resource, and weeks 1–5 complete. Pro is a one-time ₹499 ($9 outside India) that unlocks every lesson, simulation, practice set and interview question, with a 7-day refund. Can't afford it? Ask for a free pass.

Do I need a GPU?

No. Most weeks run on a laptop. For training and fine-tuning, free Google Colab or Kaggle GPUs are enough, and the plan says when you need them.

What if I fall behind?

Sundays are buffer by design, and the recovery guide says exactly what to cut and in which order. The core weeks (attention, RAG, agents, evals) are never cut. You can also re-plan from today in one click.

Is my progress saved?

Yes, in your browser: day status, hours, notes, lessons understood, quiz scores, videos watched, projects and papers. Export a backup any time; accounts with sync are on the way.

The first station is week 1. It starts with Python.

Forty weeks from now you could be shipping RAG systems and agents, with the evals to prove they work.

Start week 1 Open the app