Category
Models, embeddings, evaluation and the engineering practice around generative systems.
4 articles
Mathematics and programming often describe the same idea in two different languages.A mathematics textbook may write: A Python developer may write:y_hat = w1 * x1 + w2 * x2 + bBoth expressions…
Rajiv Kumar·Sep 16, 2026·6 min read
An AI coding session ends with a great deal of context in the conversation and none of it anywhere else. A short, structured handoff written at the end is what makes the next session start from where this one finished instead of from the repository.
Rajiv Kumar·Aug 11, 2026·11 min read
An AI assistant opening an unfamiliar repository has to infer everything, and inference produces confident wrong answers. A short brief that the assistant reads before touching anything is the cheapest quality control available.
Rajiv Kumar·Aug 11, 2026·10 min read
AI coding sessions end and take their context with them. The fix isn't a bigger context window — it's a small, disciplined set of files: what's true right now, what's already been decided, and a status marker on every claim so a plan never gets mistaken for a fact.
Rajiv Kumar·Aug 11, 2026·12 min read
