Changelog
What changed
Content and product updates, newest first.
Version 2: staff-depth rewrite
- Every term page now has day-one defaults, numbers with conditions, a decision table, cost maths, scale triggers, a failure story, what to monitor, and primary sources.
- Every interview question now shows how Mid, Senior and Staff answers differ, a follow-up ladder, red flags, a time-boxed outline and a scoring rubric.
- New Volume V — System Design & Interview Loops: 40 questions on AI system design with capacity maths, ML infrastructure, estimation and debugging drills, and behavioural rounds.
- Quizzes rewritten so answers can't be guessed from option length; every option is now explained, and misses come back in a spaced-repetition review queue.
- New: calculators (KV-cache memory, cost per request with caching, agent cost per task), cheat sheets per volume, 7-day and 1-day interview plans, progress export.
- 25 LLM-specific coding problems (tokenizer, attention, sampling, BM25, RRF, eval bootstrap, agent loop) and 5 multi-stage practical builds.
- 15 new long reads and rewrites of the original 18 with citations, real-API versions and tests.
- Corrections: temperature 0 is not a determinism switch on hosted APIs; prompt-caching savings now include the cache-write premium; lost-in-the-middle reframed for 2026 models; MCP updated to the current specification; re-index cost estimates fixed.
Launch
- Four volumes, 189 term pages, 100 interview questions, 18 posts, five level tracks and a 100-problem coding round.