Primer
How Large Language Models Work
From tokens, embeddings, and Transformers to training, RAG, tool calling, and end-to-end product flows.
20 chapters · 5 interactive demos
Notes from buy-side work: what worked, what didn't, and what we would try next.
Read noteFrom tokens, embeddings, and Transformers to training, RAG, tool calling, and end-to-end product flows.
20 chapters · 5 interactive demos
A scrollytelling landscape of global AI venture capital, mapped by concentration, geography, capital structure, and founder pedigree.
666 deals · 403 companies · 759 founders
A 2026 opportunity map built from 16+ platforms and 600+ user pain points, with 50 primary research files.
600+ pain points · 27 opportunities · 50 files
Independent notes on technology & markets, written by Yijie Ma.