The build-first path into modern deep learning: Karpathy's Zero to Hero, nanoGPT, the papers that are actually worth reading, and the engineering layer around LLMs.

The ML bookshelf that holds weight: Kevin Murphy's Probabilistic Machine Learning volumes as the spine, with ESL, Bishop, and the math that makes them readable.

The networking path: Kurose & Ross for the map, Beej for sockets, TCP/IP Illustrated for the wire, High Performance Browser Networking for the web, and the blogs where protocol engineering happens in public.

The OS path: OSTEP as the spine, xv6 and MIT 6.1810 for building, TLPI for the Linux interface, and where kernel-adjacent engineering knowledge pays off.

The resources that actually build contest skill — CSES, the CP Handbook, Codeforces practice done right, and USACO Guide — plus the practice loop that separates improvement from grinding.

A short, opinionated shelf across engineering, science, and judgment — every entry earned its place by being worth a second read.

The research and the practical playbook for skill acquisition: Ericsson's Peak, Deep Work, ultralearning, and how any of it maps to engineering.

Where to actually learn mental models — the books, the one speech that started it all, and the handful of models that earn their keep in engineering work.

The small set of sources I keep returning to for making better decisions: Farnam Street, Munger, decision journals, and the habit of inversion.

University courses with public videos and slides, collected in one table.