A field guide, not a roadmap clone
The route from first principles to the frontier.
One hand-drawn map of the whole territory — Foundations through AI Engineering, with a research track branching off the moment you can build models. Curated, ranked, sequenced: what to learn, in what order, to what depth. The walking is still years of real work; this just makes sure you never wander.
Five stages, no optional ones.
Each stage names what Competent, Expert, and Principal look like — not just a list of topics. The depth ladder runs from Aware to Principal / Researcher. The map is honest about what separates a good data scientist from a senior one.
The teal branch is the Research track. It forks from Stage 2 for anyone serious about MSc, PhD, or contributing to the literature — and it runs in parallel with the practical stages, not after them.
What each stage covers
Start free at Stages 0 and 1. Deeper stages follow in sequence.
Stage 0
FreeFoundations
The bedrock before any role. Python, Git, SQL, the command line, and a reproducible notebook you can defend.
Stage 1
FreeData Analyst
Rigorous analysis, SQL at production scale, statistical inference, and an A/B test you'd defend in a post-mortem.
Stage 2
Data Scientist
Classical ML, probabilistic thinking, experiment design, and the difference between models you build and ones you use.
Stage 3
ML Engineer
Training infrastructure, deployment, monitoring. The gap between a notebook that works and a system that ships.
Stage 4
AI Engineer
Foundation models, RAG, agents. The frontier as it stands — every volatile entry is dated and flagged for review.
Plus: Research track (branches from Stage 2) · Mathematics curriculum (Tiers 1–3, cross-linked to every stage)
What makes this different
- Ranked, not listed
- Every resource has a rank within its topic. Rank 1 is where you start; the deeper picks are for when you need more. No list of forty 'must-read' books with no ordering.
- Sequenced by dependency, not hype
- The route runs in learning-dependency order: linear algebra before ML theory, numpy before pandas, evaluation metrics before model selection. You can't skip stages — the map names exactly why.
- Honest about depth and time
- Exit criteria name precisely what 'done' looks like at each stage. Time estimates are honest ranges, not aspirational minimums. The research track is described from the outside of active research.
Every resource in one place
Books, courses, papers, and docs — ranked within each topic, with honest verdicts, free picks flagged, and no dead weight. Filter by type, level, cost, or stage.
Ready to begin?
Start free at Foundations
Stage 0 takes 4–8 weeks (2 if you already code). It has no prerequisites. Every person on the map started here.