One question, one video, one study sheet
Each explainer answers a single question in a few narrated minutes, with a study sheet to read alongside it.
2026-11-05 · 4:52Back-of-the-envelope estimation for an LLM system
Turn users into requests, requests into tokens, and tokens into dollars, GPUs, storage and latency. Use round numbers and say every assumption out loud. The interviewer is grading the chain of reasoning and what you decide from it, not whether you land on the exact figure.
2026-11-05 · 4:43Streaming, sync, async or batch: choosing how an LLM feature runs
The same model call can be served four ways. Interviewers ask this to see if you design around the person or program that is waiting, instead of sending every request down one synchronous path and hoping it finishes before the gateway times out.
2026-11-05 · 5:41Caching in LLM systems, layer by layer
There is no single "LLM cache". There are five, each skipping a different piece of work: the whole call, the input tokens, the embedding, or the database lookup. A good answer names them in the order a request meets them, says what each saves, and says how each one can serve a wrong or leaked answer
2026-11-04 · 5:42The building blocks of a production LLM app
This is the skeleton under every AI system design answer. The interviewer wants to see that you know the model is a small part of the system, can draw the rest in about three minutes, can say why each box exists, and can pick which boxes matter for this problem.
2026-11-04 · 4:41Non-functional requirements for AI systems
Availability and scalability still matter, but an AI system needs eight more targets that a normal service doesn't: quality, latency, cost, safety, freshness, reliability, explainability and determinism. They pull against each other, so you name a number for each and agree with stakeholders which on
2026-11-03 · 5:40Structuring an AI system design answer
The interview is a short simulation of designing with a senior colleague under ambiguity. There is no single correct architecture. You're judged on how you reason, and a framework keeps you from solving the wrong problem or spending all your time on your favourite part.