Skip to main content

Roadmap

How Applied AI Mastery is being built: a two-month pilot of three modules, then four release rings from foundation courses to expert seminars, with honest numbers on the size of the job.

The curriculum is 73 courses, 671 modules and about 2,684 lessons. That is far too much to release at once, so it ships in stages. Each stage is a complete, useful product on its own.

Where we are

Live now: 40 modules, 160 lessons and 40 interactives. The first of them went up on 12 November 2026.

The build began with a pilot of three modules, chosen to test the method because together they exercise every kind of figure the curriculum needs:

Module What it tests
M02.4 Matrices as linear maps Geometric figures and a drag-to-explore interactive
ML01.5 Gradient descent from scratch Step-by-step motion, code that learners run, numeric traces
RL01.6 Bellman equations Diagrams of processes, full proofs, a simulation

Ring 1 is now being written course by course. Math Foundations, Linear Algebra I, Calculus I and Machine Learning Foundations are complete, and Python from Zero and Exploratory Data Analysis are being written. Every module passes the automated checks before it goes live, and each one is re-derived line by line in an independent second check.

The pilot stage is finished when four things are true:

  1. Five learners at the target level have worked through the twelve pilot lessons, and their feedback is folded in.
  2. An independent reader has re-derived the maths in every lesson.
  3. The real cost of a lesson, in hours, has been measured. Everything after depends on that number.
  4. The style guide is frozen as version 1.0.

Until then every lesson carries an in review label.

The four rings

Ring Courses Modules Lessons Which courses
Ring 1: Foundation Courses 8 Modules 77 Lessons 308 Which courses M01, M02, M04, M06, P01, P02, S01, ML01
Ring 2: Spine Courses 12 Modules 118 Lessons 472 Which courses M03, M05, S02, ML02, ML03, DL01, DL02, DL03, DL04, DL05, RL01, RL02
Ring 3: Breadth Courses 36 Modules 327 Lessons 1,308 Which courses M07, M08, M09, P03, P04, P05, P06, P07, S03, S04, S05, S06, S07, ML04, ML05, ML07, DL06, DL08, RL03, RL04, A01, A02, A03, A04, A05, A06, A08, A09, E01, E02, E03, E05, R01, R02, R03, C01
Ring 4: Expert Courses 17 Modules 149 Lessons 596 Which courses M10, M11, M12, P08, S08, ML06, DL07, DL09, DL10, RL05, RL06, RL07, A07, E04, R04, C02, C03
All four rings Courses 73 Modules 671 Lessons 2,684
  • Ring 1, Foundation. The eight courses everyone needs first: school-level maths, linear algebra, calculus, probability, Python, scientific Python, exploratory data analysis and machine learning foundations.
  • Ring 2, Spine. The main road from those foundations to modern deep learning and reinforcement learning.
  • Ring 3, Breadth. Statistics in depth, data engineering, applied AI (language, vision, speech, recommenders, classical AI), ML engineering and responsible AI.
  • Ring 4, Expert. Research-level courses, which are re-verified every six months because the frontier moves.

A ring ships only when every lesson in it has passed review.

The size of the job, honestly

At a planning target of six hours of human work per lesson, the full curriculum is about 16,000 hours. That is eight years for one person working full time, or about two years for a team of four. The six-hour target is an assumption that the pilot will replace with a measurement.

This is why the build is in rings, why the work is open, and why contributors and reviewers matter so much. The project can pause after any ring and still be worth having.

Decisions already made

  • Every lesson is free to read, and will stay free. No paywall, no sign-up, no tracking.
  • Open source. Lessons and figures are CC BY-NC-SA 4.0; code is MIT.
  • Python only. R and Julia are out of scope.
  • English first. The licence allows translation.
  • Classical AI is in. A09 Classical AI: Search, Logic and Planning covers search, games, constraint satisfaction, logic, knowledge representation and planning.
  • Domains live in casebooks. Fields such as finance and health get a worked case in A08, not a full course.
  • Scope is the course list. A new course enters only through a written proposal.

Keeping it alive

Visual teaching material is expensive to maintain, and good projects have stalled for that reason. So upkeep is designed in:

  • Figures are code. Fixing one is a one-line change, not a redraw.
  • Every formula has a numeric test that runs on every change.
  • Lessons at L0 to L2 are re-verified every 24 months, L3 every 12 months and L4 every 6 months. A lesson that misses its date shows a banner saying so.
  • Every correction is recorded in the public changelog.

Follow the build

Back to top