take home

Reported loop detail: roughly 5 hours to build a RAG system, agent, or eval harness on OpenAI's APIs, a recorded video walkthrough, then a 60-minute defense. The hour-by-hour allocation, why the eval section is the differentiator, and what the video is actually screening for.

A working demo leaves authorship and understanding unresolved. Defend the decisions, reproduce a failure and adapt the code under a changed requirement without overstating what you tested.