What serious practitioners actually study — not a credential mill.
Curated tracks from high-school-friendly stats through production agents. Each card is an outbound link (courses, docs, books, standards). Filter by proficiency; expand a track to browse. No pay-to-rank listings.
Foundations — math, code, probability Linear intuition, basic probability, and a real programming baseline before you chase transformers.
Khan Academy — statistics & probability
Free, paced intuition for distributions, inference vocabulary, and Bayes-friendly thinking — the least painful on-ramp before ISL or ML courses.
Open resource → Beginner3Blue1Brown
Visual linear algebra and calculus series — builds the geometric mental models neural nets later assume you half-remember.
Open resource → BeginnerThe Python Tutorial
Official language tour — boring in the best way. Pair with venv + pip and a small CLI project; most ML tooling is still Python-first.
Open docs → IntermediateNumPy quickstart
Vectorized numerics and broadcasting — the substrate under PyTorch tensors and scientific stacks; worth drilling before serious DL.
Open docs →Classical machine learning Supervised learning, generalization, and honest baselines — still how most tabular and “small data” wins happen.
fast.ai — Practical Deep Learning
Top-down coding-first course family; many teams still route newcomers here for “make something work, then read the theory.”
Open course → BeginnerAndrew Ng — Machine Learning
Foundational MOOC for cost functions, regularization, and classic algorithms — still a shared vocabulary across industry.
Open course → IntermediateAn Introduction to Statistical Learning
ISL — free book + labs bridging stats and ML with R/Python; the sober complement to hype-heavy “AI” syllabi.
Open book → IntermediateHugging Face — Learn
NLP course and ecosystem tutorials — bridges classical pipelines into tokenizers, datasets, and modern model APIs.
Open hub →Deep learning & neural nets From tensors and autograd to architectures you will actually ship in vision and sequence models.
PyTorch tutorials
Official path through tensors, nn.Module, training loops, and domain recipes — default stack for research and many product teams.
Open docs → IntermediateStanford CS231n
Convolutional nets, training dynamics, and vision applications — dense but standard reference material for CV literacy.
Open course → AdvancedDeep Learning (Goodfellow et al.)
Free textbook for the full mathematical stack — best as reference and second pass after you have written real training code.
Open book → IntermediateNeural Networks: Zero to Hero
Andrej Karpathy’s from-scratch series — builds intuition for backprop and small GPT-like pieces in code, not slides only.
Open series →LLMs, prompting & RAG Context windows, retrieval, embeddings, and the glue patterns behind most “AI products” in 2025–2026.
OpenAI — prompt engineering
Vendor docs on clarity, structure, and tool use — pragmatic baseline even if you standardize on another provider later.
Open docs → IntermediateOpenAI Cookbook
Patterns for RAG, evals, function calling, and migration — short notebooks practitioners copy from and adapt.
Open cookbook → IntermediateFull Stack Deep Learning — LLM bootcamp
End-to-end LLM app concerns: deployment, monitoring, and product — not just calling `chat.completions` in a notebook.
Open course → BeginnerCohere LLM University
Gentle path through embeddings, classification, and retrieval concepts with a coherent narrative for builders.
Open track →Agents, tools & production glue Orchestration, APIs, and the reliability layer between “demo” and something finance or ops will tolerate.
LangChain tutorials
Runnable chains, tools, and RAG wiring — ecosystem moves fast; treat as patterns to steal, not a religion.
Open docs → BeginnerAnthropic — courses & education
Prompting, long context, and safety-adjacent product guidance from the lab behind Claude — good for policy-aware teams.
Open Anthropic → IntermediateOpenAI — function calling
Structured tool use from the API — the contract layer most “agents” eventually depend on for side effects and integrations.
Open docs → AdvancedModel Context Protocol
Open standard for tools, resources, and prompts between models and hosts — worth understanding if you ship IDE or agent platforms.
Open spec →Evaluation, red teaming & governance When “vibes-based QA” stops scaling — benchmarks, adversarial testing, and procurement-friendly risk language.
OWASP Top 10 for LLM Apps
Industry checklist for prompt injection, insecure output handling, and supply chain — common vocabulary for security reviews.
Open project → IntermediateNIST AI Risk Management Framework
Governance and lifecycle framing for trustworthy AI — what enterprise legal and risk teams often ask you to map to.
Open NIST → AdvancedHELM (Stanford CRFM)
Holistic evaluation lens for models — transparency-oriented benchmarking beyond single leaderboard scores.
Open HELM → AdvancedOpenAI Evals
Open framework for model-graded and deterministic evals — patterns for turning fuzzy product requirements into test suites.
Open repo →MLOps & delivery Versioning, CI/CD for models, observability, and the boring wins that keep GPUs from becoming expensive science projects.
Made With ML
End-to-end ML design, deployment, and testing lessons — strong bridge from notebooks to something resembling software engineering.
Open site → IntermediateMLOps Zoomcamp
Free cohort-style curriculum from DataTalks Club — experiment tracking, deployment, and monitoring with real homework.
Open repo → AdvancedKubeflow documentation
Kubernetes-native ML workflows — heavier than a single script, but the shape of ML platforms at larger orgs.
Open docs → BeginnerDocker guides
Containers are the packaging layer for most inference services — worth solidifying before Kubernetes or cloud-specific courses.
Open guides →Data engineering & vector retrieval Chunking, freshness, metadata, and ANN indexes — where RAG quality is actually won or lost.
Pinecone learning hub
Vector search concepts and RAG architecture explainers — vendor-tied but readable introductions to embeddings workflows.
Open hub → AdvancedDesigning Machine Learning Systems (Chip Huyen)
Book site and companion material on data loops, deployment, and evolving models in production — senior IC interview canon.
Open book → IntermediateApache Airflow documentation
DAG-based orchestration for pipelines that feed features and batch jobs — still a default answer for “how do we schedule this?”
Open docs → IntermediateSnowflake tutorials
Warehouse-centric patterns for governed data — where many enterprises centralize features and audit trails for analytics + ML.
Open tutorials →Logos via public favicon service for identification only; trademarks belong to their owners. Curated for orientation — verify pricing, terms, and prerequisites on each provider site.