Multi-armed and contextual bandits as adaptive decision-making: the explore-exploit trade-off, ε-greedy, UCB, and Thompson sampling, and the inference and safety costs that adaptive traffic allocation quietly introduces.

What changes when clean random assignment is unavailable: the identification assumptions, estimands, and diagnostics behind Difference-in-Differences, Propensity Score Matching, and Instrumental Variables.

Retrieval-Augmented Generation (RAG) has rapidly become a widely used architecture for bridging the gap between static Large Language Models (LLMs) and dynamic, proprietary data. By fetching releva...

KV Cache: How LLMs Avoid Recomputing the Past Large language models generate text one token at a time. At every step, the model attends to all previous tokens. Naively, this would require recomput...

Anyone who has played with a Large Language Model (LLM) API has likely tweaked the temperature slider. The conventional wisdom is simple: “Low temperature = boring and factual, High temperature = c...

In the previous posts, we explored how an ad impression is auctioned in milliseconds and how machine learning models decide whether and how much to bid. However, every system we’ve discussed - from...

In the previous posts, we explored the plumbing of AdTech: how ad tags, redirects, pixels, and browsers coordinate to make a single impression possible. But for a data scientist or engineer, the mo...

In the previous posts, we explored who participates in the AdTech ecosystem and how the supply and demand stacks are structured. In this post, we shift focus to how those systems actually talk to e...

In our first post, we looked at the high-speed “Invisible Auction” that happens every time a webpage loads. In this post, we’ll move from the marketplace overview to a single question that turns ou...

Welcome to the first part of our AdTech primer. For most developers, modern advertising technology is a black box of invisible network requests. In this post, we’re cracking it open to understand t...