1,231 saved links with two distinct personalities: a serious mathematics self-study library (you hoard linear-algebra texts the way others hoard novels), and the practical stream of AI-era data science — where the job is quietly being redefined around evals, agents, and semantic layers.
Related: Machine Learning · Data Engineering · Python · LLMs
Start here
- Maxime Labonne’s LLM course roadmap — the modern data scientist’s curriculum.
- 3Blue1Brown’s lessons — “hands down one of the best visualizations of how LLMs actually work” (your own note).
- Seeing Theory — interactive statistics from Brown — foundations matter more in the AI age, not less.
- how LLMs are built: pre-training to post-training and understanding LLMs from scratch, using middle-school math.
- what a data-analyst portfolio actually looks like.
The mathematics shelf
The densest self-study collection in your entire vault — mostly free, university-grade.
- Linear algebra: Halmos’s Linear Algebra Problem Book (“a complete inquiry-based course”) · Terence Tao’s lecture notes · Linear Algebra Done Right — free electronic edition · The Art of Linear Algebra — graphic notes · A Rough Guide to Linear Algebra · a beginner-friendly full course · mecmath.net.
- The big compendiums: MIT’s Mathematics for Computer Science — the 1,048-page PDF · Mathematics for Machine Learning · Imperial’s Mathematics for Inference and Machine Learning · a maths, CS & AI compendium · “a math degree is 1000x more useful than a CS degree” — the repo to back it · OpenStax free textbooks.
- Deeper cuts: Harvard’s Advanced Calculus (Loomis & Sternberg) · Sussman & Wisdom: Functional Differential Geometry · Kowalski’s linear algebra texts · The Complete Mathematics of Neural Networks and Deep Learning (video) · the maths you need to survive AI.
- Statistics: The Art of Statistics · Head First Statistics — free PDF · building statistical literacy.
- Reading habit: the math-blog canon you saved in one sitting — terrytao.wordpress.com · johndcook.com — blog · quantamagazine.org · “Think in Math. Write in Code.”.
Data science in the LLM era
The discipline’s new centre of gravity: evaluating systems that produce text instead of numbers.
- evaluating LLMs for inference — one size does not fit all.
- LLM-as-a-Judge: a practical guide · building a judge that aligns with human judgment · LLMs for automatic evaluations.
- evaluating LLM systems: metrics, challenges, best practices · methodologies by language task · hands-on with DeepEval.
- why GenAI app quality often sucks — a framework and “evals are not all you need — it’s a data science and monitoring problem”.
- five revelations from “Why Do Language Models Hallucinate?”.
- measuring LLM reliability in a RAG system · hands-on LLM monitoring and observability.
Agents for analytics
Where your data and agent interests meet — the analyst’s job, delegated.
- a data-analytics agent with LangChain and DuckDB · pandas reports with local models — no data leaves the machine.
- a CSV sanity-check agent — LLMs can do repeatable, deterministic tasks.
- a multi-agent SQL assistant with human-in-the-loop checkpoints and cost control.
- Google’s Data Agent Kit — open-source data engineering & science skills · automating data cleaning with AI · optimizing a data-analysis agent with GEPA and execution feedback.
- “this whole MCP thing will change data analytics as we know it”.
- The Samir Saci supply-chain series: n8n agents for supply-chain analytics · agents + optimisation algorithms · the portfolio version.
Semantic layers & Fabric
The enterprise data stack you work in — and the ontology thread that runs through your architecture thinking.
- “Your AI is guessing — here’s how to make it stop”: ontologies, knowledge graphs, and a semantic layer.
- ontology and graph databases · Microsoft’s Ontology Playground (preview).
- Microsoft Fabric interactive exercises · Fabric Data Agent over structured and unstructured data · the Fabric DataOps sample.
- Semantic models in practice: semantic models for analytics, reporting, and AI · documenting Power BI semantic models with Fabric data agents · auto-populating data agents with semantic-model synonyms.
Career & craft
- “You’re never too old to chase your childhood dreams” — an AI career journey — decades of wisdom on mentors and continuous learning.
- the evolving role of data scientists in the age of intelligent automation and AI as complement, not replacement.
- from school to work: error handling and production Python · better unit tests, in small steps.
- 50 real-world datasets for projects · basecs — hand-drawn data structures and algorithms.
- storytelling + design for unforgettable presentations — the last mile of every analysis.