The Project Deep Dive: Five Layers of Follow-Up, Traces of the Choice, and Where Every Number Came From
The project deep dive carries the most weight of any part of the interview, and it judges exactly one thing: whether you actually did the work. Today the deep dive becomes five layers of follow-up, and you practice leaving traces of the choice at every layer. Then real numbers measured in several courses on this site show how to talk honestly about a course project. The interviewer agent learns to catch claims that cannot be checked: percentages without a denominator, phrases like significantly improved, a story told entirely in we, and numbers nobody can say how they were measured.
Today's Goals
- Can prepare a project with five layers of follow-up, what you did, why, why not the alternative, where the numbers came from, and what you would change, and name the rejected option at every layer
- Can talk honestly about a course project: which part you changed yourself, the denominator and baseline behind each number, and what it does not show
- Can implement a claim checker that catches four kinds of uncheckable claims and generates a follow-up for each, without flagging real numbers that carry a denominator
The first three days practiced what you know. Today practices what you have done. When you finish, come back to the top and check off the three goals.
Plain-Language Walkthrough
Inspecting a used car: the seller's word counts for nothing, the service records and odometer do
Anyone who has bought a used car knows the seller's car is always "in great shape, never in an accident, always serviced at the dealer". An experienced buyer does not argue. They do three things: go through the service records and check that dates and mileage line up; read the odometer and compare it with the last service entry; open the hood and look for signs of work. They do not judge whether the seller is honest. They judge whether each statement can be checked.
The project deep dive is the interviewer inspecting your car. You say you "built an enterprise agent platform and improved accuracy by 40%". They will not call it a lie. They will just follow up, layer by layer: which part of the platform did you write? What is the denominator of that 40%? Compared with what? Why not the other approach? Each question turns another page of the service record.
Day 28 of the 30-day course put it this way: this part judges one thing only, whether you actually did the work. The way to tell is not to listen to what you did, but to listen for whether you can say why you did not do it another way. Today turns that into things you can prepare: five layers of follow-up, traces of the choice, honest numbers, and I versus we.
One thing first: preparing for the deep dive does not mean preparing a script. You can find plenty of "ready to copy" project stories and interview lines online. Memorize one and you pass the first layer and collapse at the second, because you cannot name the path that was rejected. Everything taught today assumes the thing really happened, even if it was only a course exercise.
Interview questions
Tell me about the agent project that best represents your work.讲一个你最有代表性的 Agent 项目。
Common in ChinaCommon overseasIntermediate#project-deep-dive#storytelling🔒 Unlock this chapter to see the answer keys.
What is the most important number in your project, and how did you measure it?你项目里最关键的一个数字,是怎么测出来的?
Common in ChinaCommon overseasIntermediate#metrics#evaluation#project-deep-dive🔒 Unlock this chapter to see the answer keys.
Why did you choose this approach over the alternative?当时为什么选这个方案,而不是另一个?
Common in ChinaCommon overseasDeep dive#tradeoff#project-deep-dive#decision-making🔒 Unlock this chapter to see the answer keys.
If you rebuilt this project, what would you change?如果让你重做这个项目,你会改什么?
Common in ChinaCommon overseasBasic#retrospective#project-deep-dive🔒 Unlock this chapter to see the answer keys.
