The Atlas series
Each Atlas takes one field and works through it in dependency order — why a thing exists, how it actually works, and what breaks once it meets real traffic. Written to be studied, not skimmed.
Architectural Evolution, 2017–2026
Ten years of architectural churn, in dependency order: what a language model actually is, what the Transformer replaced and why, the lineages that descend from it, and the economics that decide what gets built.
Building systems on top of models you did not train
The application layer, in dependency order: embeddings and search, vector stores, RAG, agents and tool use, MCP, and everything that sits between a working demo and a system that survives real traffic.
Data structures and algorithms, derived rather than memorised
The structures and algorithms interviews ask about and systems actually use — built up from why each exists and what it costs, with the trade-off that motivates it made explicit.
Being written
From statistics to deep learning, with the maths kept in
Statistical foundations through classical models to deep learning — the mathematics that justifies each method, the code that implements it, and worked examples on real data.
Being written