Crack Analytics Interviews with an AI-Powered Prep System — Data Neighbor Live

0:00 uh please brace yourself for a lot of back and forth between these slides and some of the cloud code and other AI workflows that I'm going to share with you today. Okay, what are we going to chat about? We going to chat about acing analytics interviews with AI. Actually, it'd be pretty cool in the chat if you all can share which domain are you from. Are you from data science, data an analyst or product or even maybe you know uh design if if there are some that would be pretty cool to know. Uh and hi Aran let me know uh I can't see the screen as much I'm sharing my screen.

0:34 Yeah, we got a lot of product got some data. We had a sales product analytics. >> Oh, nice. >> Science engineering. >> Yeah, that's pretty cool. >> A lot of product data science. Yeah. >> Nice. Nice. So, whichever domain you are guys, you can imagine you this. Yes, we're going to talk about interviews specifically, but obviously uh interviews do test your analytical skills and which means learning how to deal with data. So anyone and everyone who looks at data is probably going to get benefit from this session or at least that's a hope. Okay. So let's get started. Before we move on, let's have a quick question to you all. Let's say your company launched a free trial. This is a question that we it's a typical question that we get in interview probably product u an analytics interviews. Your company launched a free trial for its premium t tier for 3 months. Um I mean 3 months ago signups are 40% above target revenue is 50% below target. What would you do? Let's say this was the question asked to you guys. Um can you choose which one you would choose in A, B, C, or D? Here we kind of gave you some hints about why you'd pick it. Probably this is a different style I'm drawing the time, but it would be cool to know. Got a lot of bees. A lot of bees coming in.

2:00 >> That's nice. That's nice. Okay. So, >> I also had another question for you, Stravia. Is this I had a question from >> is this just a is this just for helping us with like analytics questions or also with like product sense questions? That's a great uh so th this is probably going to be uh slightly more towards analytics but all the material that we are going to talk about how we generated with cloud code and AI it would it could absolutely be used even for product sense. So um and you know the the lightning lesson is in the lie of a course that we have around a analytics for builders and that is absolutely something that you could u understand more from like analytics standpoint and what we uh share as part of that work would help you with product sense as well. So maybe we could see uh somewhere to the end of the session. I'd love to know for the product sense folks if how much you thought this was beneficial but uh I would say it's slightly leaning more towards analytics. Okay, cool. Okay, looks like we do have a good chunk of folk here who would go with clarifying questions which is the right answer and I wanted to set you up for success guys. Uh so great that you chose this. Uh now let me before going into the major workflows or you know the uh lightning lesson I want to basically share with you what AI can build and how it can give you the right frameworks that set you up for success for interviews in this space. Okay, bear with me while I share my screen.

3:42 And hi I'm Sean. Please let me know if there are any questions in regards to what I'm sharing. You can pause me anytime. Okay. So today's interview uh we are choosing Door Dash. Door Dash for people who are not aware from other parts of the world. I think it does have u like affiliates in Europe but Door Dash is a um like you know uh marketplace company where you have three-sided marketplace where they're dashers, consumers and merchants. It's a food u you know delivery and food pickup service. It's a pretty popular um like you know I think for people from India Swiggy and Zumato are one of them. So we picking door dash today. So this is the final output that I have on a plan of uh let's say you're a staff data scientist and this is your interview prep brief. This is the company intel uh and it took these as the as the metrics.

4:41 If you're a staff data scientist, if you're a senior, and if your strength is metrics, and your gap is experimentation, and it's a three-sided marketplace, this would be like a prep plan for you, a interview prep brief. Metric trees. This I would say is something that's absolutely uh like at least the first time I saw it, I was in love with these metric trees. What uh is the top metric and what all metrics contribute to it. So I thought uh and not just that metric you have for every other important metrics and uh like what are the guardrails for each of them and stuff and then for experimentation what type of experimentation happens like when do you do this? This is almost like a brief like a cheat sheet talk on what claude code or AI helped you to understand how you want to go about you know preparing for your interview. And this is like a 7-day plan. Do this on day one. Do this on day two. Do this on day three. Do this on day four and five.

5:43 And you know, communication rules, practice, and a rubric on how it goes about it. So this is not the final one. The there are a lot of details that it helps you with. These are a bunch of docs. It basically creates a bunch of skills and agents, research agents that goes and you know helps you mine data from internet. um that's publicly available about Door Dash and helps you create these things. The Door Dash earning reports and public metric analysis uh like a deep dive about a staff PS interview. What is uniquely about it? Uh I took here uh I took uh staff data scientist as my uh you know trigger to generate these things but you could generate them with any uh title that you would want specifically in the analytics area. um door dash internal metrics what are the notar metrics what are their drivers look at this really cool like you know metric trees are it generated and a bunch of these files we'll go into the depth of it at the end but I wanted to share you all these things that like on the day of interview what's a cheat sheet like all of these were generated by claude code to help you with prep and we finally have also copilot that works with you to do mock interviews So, this is what we'll be sharing in the end part of the demo, but I wanted to share with you a flavor of what's coming next.

7:14 Awesome. Okay. So, let me go back to my presentation, get started. Okay. So, now that you saw what can AI build for you, I would probably break it into four parts. It's going to basically help you understand uh like help you take where you are today. What does the interview have? How many days you are into the interview and what are the strengths and weaknesses according to who you are. And once you share with them, what it does, it basically the first thing that it does is goes and runs a bunch of skills and agents to build these uh you know u research files, deep research. She does lead resist multiple areas and build these files for you to help you prep.

8:03 And as soon as the prep material is in, it helps you plan for how do you want to go and you know uh like you know prep like when do you read what doc, how many days do you have, is this too much of a prep or is this too little you know the stuff like that. It helps you pace yourself towards the interview date and as part of that the most important thing is helps you practice actually have a mock interview. um you know uh it's a claude project that I have right now but helps you actually go through the uh the practice sessions take the feedback and help you change your plan again so this is the big loop of the entire interview process that you could work with AI to build this okay so now how was this built whatever I shared with you how did I go ahead and build this right so everything you just saw was generated using frameworks from our Maven course. So we have a Maven course that's coming up literally in 5 days. So it's called AI analytics for builders. So what did we do there? We actually have these five uh you know different weeks uh like different uh topics that we share as part of the course. Question framing analytics tools and workflows metrics and root root cause analysis experiment design and analysis storytelling and communication.

9:21 So it takes all of this u the sixth bonus week guys as part of the course we'll have an interview prep toolkit everything that I'm sharing today would be part of that as well. So it took all of these things and it condensed into all these files that it prepared. Basically it structured how you scope any question. It powers the agents that build the prep on how it's taking what analytical uh you know uh workflows that needs to be questioned for door dash interview per se and then creates the metric decompositions based on what is the right way of doing the metric trees.

9:55 So we talk about metric trees and metric root cause. So it uses that information to create this for your interview prep um and then generates experiment specific questions because we have a big giant experiment design and analysis section in the inter in the course and then it teaches how to answer structure and framework. So, how I got to where I got with those prep files was because I was sitting on this giant re repo of the like, you know, really rich content from the uh course and it helped me bring you know those uh like documents into like reality.

10:34 That said, I do have some giveaways for you because you signed up and you you you are like listening to us either live or you know on video. I do have something that I could share with you that you could use today um for if for some reason you can't make it to this course. So a quick plugin about the course. One very cool thing that we do with the course is we give you a repo called air analyst plus. It's contains 43 skills, 29 agents and you know so many other things just uh and it helps you from question framing to storytelling the entire thing. We actually run a boot camp as well to help you understand the details of all of this. We share uh some of this as part of our u uh course as well. But you get this entire repo where you could take this repo and you know plug in into your company and you know work with it probably. So you would also get the six week uh you know bonus course is the uh bonus course that we have because if you see it's a fiveweek one. It's about the interview copilot.

11:35 It helps you, you know, coach, practice, demo, simulate any mock interview that you have. And you also have an interview prep pipeline. There's seven skills and 21 agents that helps you create these documents and details for yourself. By the way, before we move on with the rest, I want to let you know that we have a special discount for you guys who are attending this course. It's called clot 30. You get a $540 off. Uh this is our first ever course in this area. We did a boot camp one. It was highly successful. We got 4.9 rating on that. Uh and we aim to deliver the same amount of value in analytics for builders. So please check this out. And if you're interested, you could use this code to get uh this off.

12:21 Okay. So now uh let's talk about what interviewers actually score you on today. Right. So most uh candidates think interviews like test knowledge. I mean it's not wrong they do test you on knowledge but what they test you on I would think it's majorly about structured thinking. I myself I think I've conducted at least hundreds of hiring management interviews. I look at these scorecards from uh data scientists from my team and I and all of the really good candidates were technically amazing. uh what they lack is the structured way of approaching a problem right so whether so these like you know dimensions of structured thinking the scope the metric the decomposition the trade-offs the decision I think these are in these areas what people generally miss is they just jump straight into answers now they do ask some clarifying questions but just for the sake of it because they are told to ask and not get into the depths of The reasoning was the exact metrics to go after and you know they kind of skip the decomposition part of what a metric the metric trees cover that right like let's say if you talk about revenue revenue u is not a number just by itself it is a combination of so many other like important metrics for example users into conversion into arpo so knowing this detail and also communicating that detail in that uh you know um structured is what is basically going to help you in these analytics interviews and giving one-sided recommendations with like not looking at ROI and cost analysis and also most of the people always miss the decision like ending with what is the clear and specific recommendation that you have. All of these things I think helps you get a better score for your interview. And this the rubric gets tougher and tougher as you are uh you know interviewing to more senior roles versus junior roles. Structured thinking just gets e the highest priority as you get into the senior roles.

14:36 Okay, let's think um let's I mean I have a bad example and then good example for you guys. Looks like we have a very good crowd today. uh probably you would go for the good analysis good example but just in case and wanted to share like what a bad example looks like right so remember like uh if you have this question the the question that I asked let's say you didn't clarify which product which tier what time window you basically did like a basic V1 like a first level of clarification and you did not go into the depth of different layers of clarification that is also like an norm that oh this person's not thinking through this right and then regarding metrics so understanding like the detail of what that metric actually means like understanding its decomposition understanding what are the tradeoffs um before like you know I'm talking about like cannibalization these are some things that people like look for when we are grading for like how good the answers were for for that interview right and especially the decision like I said most people often miss that like optimize the what and when and how do you want to like you know give a recommendation to the VP of product or you know to the like VP of engineering around hey this is what we saw and this is why it's uh not as important uh but I would like for us to go to this and this ensuring that end you end the loop with the decision is like the most important thing for example let's look at a 5x5 answer structure. So understanding the scope is it basically the total revenue or per trial uh revenue total revenue or per trial user like what is this right?

16:23 understanding more or clarifying more and then understanding like what is the 40% sign up lift from is it organic or paid because each of them would lead to an entirely different set of hypothesis and you basically want to know all the details about what the question is about before you dive into any of the next layers into the interview. Okay, metrics. I think I I shared this with you in the revenue one. So what is uh revenue specifically to that question.

16:56 So number of trials starts the conversion of those trials because not trials probably especially they're going to be free, right? You're not going to get money out of it right away. So the conversion aspect of trial is the most important one and you don't want to miss that. So trial starts into conversion into average revenue per user. you talk about this that in your interview basically gives the interviewer an idea how you're thinking how you're structuring and um like this like I said this gets more and more important as you grow in your career into staff or senior levels decomposition like lower intent users flooding in check signup source mix like trial not showing the value check feature adoption the first seven days pricing friction check checkout abandonment. These are the multiple you know hypothesis that you could generate from that metric that you decompose into like and you know you could generate each of these hypothesis and I don't have all the detail here but you basically talk to your interviewer confirm with the interviewer and ask them hey these are a bunch of things we could go into which direction do you want me to go in right and that basically tells the interviewer that yeah this person really thought through all the details and they are asking me to you know for for this interviewer's sake which area do you want to uh get into and then you most important thing is also talking about trade-off because it's not always one way or the other you basically ensure that uh you get a product out it could be cannibalizing another product you get a product out get a product feature out it could be a lot of cost and engineering and the ROI is probably not high so what are these trade-offs and ensuring that you talk about them the one after the other is what that will help you right in these interviews and the other the final most important thing is the decision like I already said. So basically this answer gets the I mean this is trying to condense the entire interview but this answer if you cover these areas would definitely get you to a good structure way beyond knowledge and if the co-pilot that I'm going to present with you later is going to score you based on how you did in all these areas and let's say you didn't do well it'll basically help you by giving you feedback and you process that feedback and come back again and try to see how you're doing for your next levels of, you know, mock interviews that you do with the co-pilot.

19:37 Okay. Uh, I've been talking a lot. Any questions or anything? Hi or Sean in the com in the chat that you'd want to raise before I move on? I >> think we can move on. >> Okay, cool. Awesome. Okay, so I just categorized into the top five type of questions. I know I talked about data science here, guys. I know we do have a bunch of product people like uh I know that I've done a lot of product manager analytics interviews as well and they're pretty similar. Uh I would say data science you expect them getting into more of statistical part in case we get into a technical depth like a causal inference. But at the higher level of analytics, I would probably have similar style for even product folks. Uh yeah.

20:25 So like uh like I said though it says yes, I think this is something that even product folk can definitely um you know utilize. So what are the type of questions that we see here? Um measuring success basically how would you measure if a dashpass is working or not? Right? uh like I said we are using door dash as an example for most of our questions here and u this like each area type of questions gives you an example of what could be this types example for door dash as a company okay and then diagnosis uh let's say orders dropped 15% last week why did they drop uh is one type of question and the other type of question is experiment design like design an AB test for this checkout flow and launching or not is basically should we expand or dash to rural areas is one example of that and for improving how would you improve dasher retention that's like you know uh one style of uh like an interview question that you get asked and one other thing I want to say is uh there is a possibility of an interview covering like more than one of these type or in some interviews all these could be a possibility too basically you start uh the question could be around hey we are thinking about launching this new feature how do you want to go ahead and measure success. So you talk about metrics and then you talk about okay now that we have these metrics that you want to measure success with how do you go about u like you know designing an experiment and once the experiment they'll give you like you know a fake some experiment result data and they would be like now that you have these results how would you evaluate these experiment results and can you give a decision of launch or not and after that could be hey now that we did what we could with the first design. What could we do with the next layer uh of experiments we could run in the same space? So there is basically a possibility of combining different types of these into like one single thing uh interview as well and like I said always you know classify the question and know which dimensions to hit and then then structure your answer.

22:42 Okay, >> we got a question uh from Vishnu. Uh uh if someone were to use this tool to prepare for use cases product sense AC so I guess C so for for example like across different domains like healthcare finance uh would it still cover those use cases? >> Absolutely. So today I'm giving you an example of Door Dash to make it easy for you to land these things better. But the tools that we're going to provide there going to be two sets of tools. What is giveaway for free today? Because you attended the sign up. You came in and you're listening to us. Um you would get something. It's a free u uh like giveaway for you guys. If you join our course, we have a sixth week bonus that we're going to share with you. You give it any company that you want to join. uh you give it the type of role you want to join and you give it uh the levels of experience you have and it's going to basically tailor the prep documents for you. It's going to basically tailor u what you need to read uh b how to how to prep for it's going to tailor the plan for you how do you want to tackle the interview and it's also going to help you with the uh mock interviews that you could do it's an entire system almost I mean to be honest that could be a course like a fiveweek course by itself uh but we are trying to uh get as much uh you know information packed into this and share it with you as part of bonus bonus that will be uh something that is paid which will be part of the course. If you take the course you get that bonus you know as well. So you get something for free today and you get something if you join the course and both of those would be some you able to cover any domain or area that you uh that you want to go after. Okay cool. Yeah. So how did we get to this 5x5 uh structure? Basically you you can see a pattern following through here. Uh that's the reason why we built the analytics for builder course the way we built it right because those are the most important things in analytics today. Wherever you are you could be a data person like actually in as a data role or you could be like a product person working with data. You could be an working with data. Uh you could be like a designer who probably or a researcher working with data. These are like different stages of uh like you know uh type of workflows you work with um when you work with data. So all of those are part of each week of our uh course structure and this is what we cover. So like I said because you're part of this course uh part of this lightning lesson today the free workshop you get 30% off which is $540 off guys.

25:33 So, it's pretty uh uh great offer right there. So, uh please um if you're interested, leverage this opportunity of this discount and we'd love to see you in the paid course. Okay. So, how much time do we have? 12:30. Uh we originally had this for 45 minutes. Um but we changed it to 1 hour, I think, Sean. So, I have 30 more minutes. What I'll try to do in this 30 more minutes is go through with you how I built this system to some extent.

26:07 uh what goals maybe maybe how I built is like an entire course the boot camp that we run but probably I'll share with you the skeleton of what goes into this uh interview uh prep ecosystem that we built and that we'll share with you right I'll share that with you uh and then probably I'll try to um share all type of documents I briefly went over with you in my first uh like a flash demo let's get into more detail there and Uh I'll share with you the co-pilot.

26:39 Uh some of those instructions are something that I'll give away uh to you guys because you're uh either watching this live or like uh in a recording because you signed up. And at the end last 15 minutes probably or maybe 20 minutes. I'll try to see if we can cover this in 10 minutes. We'll have for questions. >> Thanks Rob. I have one one question for you and you're you're probably going to answer right now because I think you're going to teach everyone how they can kind of build it for themselves. Uh but from Riy uh did you do this use claude code for this or did you use like uh yeah claude project for this?

27:14 >> Okay I use claw code guys we kind of are smitten both Sean hi and us we are like today morning claude was not working and I was like hey it's not working for me is it not working for you guys like we can't work without claude now. So anyway, uh shorter answer, yes, I use cloud hold to build these things. Uh but I gen I have a free version of this that you could use in your cloud web if for some reason you're you know you're not comfortable with cloud code yet or something. Um and um so do not worry I'll try to give uh something for you guys if you don't work with cloud code but I absolutely absolutely like suggest you to check out cloud code we have a bunch of free resources guys um it's something that I'll share with you later but since uh I got the opportunity I'll share with you we have a free claude core repo that you could use today to get started it'll pro it'll definitely help you get started to do some if you want to try it out, we have a boot camp coming later in May that where where we will try to handhold you and you know help you with getting on board into cloudboard and all of that too. Okay, so we don't have as much time. Let's jump in into how we built this and what goes into it.

28:36 Where's where's the gravity? Okay. So, what you're looking at is my anti-gravity screen. So, I'm using uh I'm using uh anti-gravity here. This is my ID. You could see on the left. This is our analytics for builders repel. This is something Sean Hai and I we run all our stuff in. And we are in lightning lesson 9 today dealing with interview prep with AI. So this is something that I already had uh looked at. So what does it do? It basically reads all my skill files in this lightning lesson that I prepared and it generates an asy architecture of the diagram showing the entire pipeline of how we built this. Let's see.

29:28 Live demo is always very interesting. It doesn't come out sometimes the way you want it, but we'll see if this one does or not. Um I was briefly sharing with you right in my uh demo the architecture behind the magic. So it's currently going through it live going through all the uh skills and agents that we have like you could see a bunch of these things on the side. It's going through that and creating it going to create this like you know as architecture for you. Okay let's see if it's actually creating.

30:04 It's no okay so it's so these are a the bunch of skills and agents that basically we did uh company intel builder so this is a skill this someone just asked right Vishnu or someone about the company and role so this is where you give it you basically give the company intelligence so it researches the business model revenue competitive landscape data team focus and all of that and it does uh and these are the three agents that this skill calls earnings researcher, blog synthesizer, competitive analyst and it basically takes the uh input whatever you gave it and and creates these things.

30:45 Okay, cool. So I was thinking it's not giving me the ask architecture but it did give me that's cool. So it basically was giving me in text format what all it did. So the metric tree builder one it basically you could see this here right? So stage one is the company intel builder because it also does uh it in a way that it uses the first stages uh input into the next few stages as well. Uh because um that's the reason company intel builder comes first earnings researcher, blog synthesizer, competitive analyst. Um this uh system does really well for companies that has a lot of information available outside.

31:25 For example, Door Dash is very good with its uh blogs. uh it has a very good data blog, it has a very good ML blog and hence I have a rich source of information that I could u you know do to help get this information for example if the company that you're interviewing for doesn't have this information right there is also a parallel uh the company knowing what company it is it basically searches for the business models in and around that company's research area for example healthcare let's say it's a niche company that you're going for but it does go and uh look at the research from the companies like the company that you're going for in that area. For now, uh Door Dash has a lot of stuff about it uh available in public. So, this is the system that it created for me to build those thorough prep documents that we used to create mob interviews. Cool. So, this is the first one and let's this is the stage two uh in parallel. It basically runs these two in parallel apparently the metric tree builder and experimentation researcher and what all things that go around it and these are the stage three trunks framework of cheat sheets um I'm a big cheat sheet uh believer I used to do that as part of my master's uh like bachelors as soon as uh I learned something put that in a cheat sheet try to look at it uh especially with our busy lives right now these are very useful and next comes stage for uh this is like a cool thing that I do for HTML. You don't probably need it if you think you have a plan already set up, but I generally uh think that's a easier way to extract all the information from these files into something simple to start with and gro and take that information in so that you could use it for um once you get the basic idea of what all are available then you can go and double click on each of them to learn more and then you have the final deliverables. It says markdown files but you could always create PDF out of it.

33:27 uh it's going to have 12 files and you know all of these things. Okay, so this is what we have. Let's try something. Let's ask about uh ASKI architecture for metric trees metric trees that you built. Let's see if it what it generates. But yeah, so what I'm asking you to do is go through one of the files. Um, we are used to at least I'm used to reading through like PDFs or printed files or something. But if you have uh all the time uh not the time if you're comfortable with working with cloud code directly, you could literally do all your learning on cloud code itself guys like I absolutely like recommend you do this. Look at this. It basically gives me what is the northstar the gross order value total order values into average order value and what is total order active consumers order frequency. It also gives you these are the numbers that it pulled from the recent earning report. Some of them that are available outside some of them that are not. Um and yeah it basically goes and does this decomposition of these active use uh like again within active consumers there's so much more right you have new users retained users churned users this is the growth uh you know uh thing right here the entire growth equation uh what each new users have organic paid referral program retained users the dash pass subscribers non-dash pass subscribers win impact campaigns. If you know about Door Dash, you would understand that it got to as detailed as it can get. Uh this is like really powerful. If we knew all this information and go with this depth of information to an interview, uh I think the chances of you know winning is pretty high or the chances of like actually feeling the interview.

35:30 Okay. So we have this. So I just shared what did I do? I shared with you the system that built this prep material, right? And I also shared with you the uh content of what goes into one of the prep material it built to get you to understand that hey, you could go read the prep file, but you could also work directly with claude code to understand and go through the information that's in this file. Right? Let me stop sharing and go to the prep docs themselves.

36:04 Where are we? Okay, so remember I was briefly going through this. Uh now let's get into the details guys. I was absolutely impressed by this. Where are we man? I have so many windows system. Yeah, I think I'm sharing the right one. Where are we? 12:40. Uh Sean and hi. Let me know when I'm going too much. I'll try to wrap this up in like five maybe 5 to 10 minutes.

36:37 >> Cool. You're good. >> Okay. Awesome. So, look at what it did. Um, Door Dash earning reports and public metric analysis, quarterly performance summary 2025. It went through the entire thing. It gave me what metrics Door Dash reports publicly, marketplace, GOV, cross order value. It gives you the formula. It goes beyond what's obviously available. This is something that it it helped you understand the revenue adjusted EITA like so these are some things that probably a financial finance analyst who could go to door dash can also get benefited by right because I'm sure these are more like terms specific to uh your the area and how Door Dash talks about business these things are very important when you get uh question when you're getting hired into higher roles strategic roles or you know uh staff roles or like a manager roles. These gets like even more um important because you not only know about the area the data domain area or the product area but you also know about how the company is doing and how the company is communicating outside and how it wants to talk about its business. So my bar it just goes into like a full depth here and then a deep dive into a stats interview. It talks about network effects very important aspect of Door Dash. So if you're not going to be Door Dash, if it's if your interview is going to be into another space, uh let's say probably maybe healthcare, maybe there's a lot of causal inference happening there and that's it's going to talk about that. But for Door Dash, this is what it created.

38:16 It's like I mean it's look at this how many it basically created you a a 16page file and how do you talk about this but another thing that I really loved about this is this is not just going to help you for your interviews this is just going to help you do a better job at wherever you are as well right because it helps you understand how big tech is actually approaching data and probably there are lots of things in this that you could incorporate in your current role, right?

38:51 So though we're going to talk this is like an interview course how um like as part of your research of different companies and domains, it's just going to make you a a very like you know strong analytics person in your own company. So this is the brief uh representation that we saw in cloud code like look at this the total revenue piece it how many it created I think metric trees full decomposition it created a metric tree for every important area revenue deosition delivery time look at this total delivery time order placement to dasher assignment order processing time dasher matching time available dasher identity in the Dash accept rate batching decision. Wow. So like look at the detail here. Wait at the restaurant and the restaurant prep time accuracy of Door Dash's prep time prediction.

39:48 Restaurant order queue. Like you take any area. It just covers the full detail of the area. And this is the reason why the interview copilot that we build on top of these documents is going to serve you as an actual interviewer because it has so much information uh like a staff data scientist or a staff product manager in Door Dash that it's going to help you be very honest with how your answers are and it grades you well.

40:18 Okay. So I can go on. These are a bunch of files that I generated. So it's all at Jessica Lax who's chief analytical officer at Door Dash. Uh she did a lot of interviews podcast and talks about her team and it talks about what the leader cares about. Yeah. So many uh like details here that you could prep but let me just do this quickly. So this is the co-pilot that you could build. I'm going to share you a mini version of this cop pilot, the max version of this with all the documents that we created, all the prep and adding into that and the instruction that you add. Look at the instruction. Uh I called it a DSM product case interview co-ilot. You could also call it a PM pro product case interview co-pilot. And you have a deep knowledge of all of these things. What different modes you have. You have multiple modes. You have a practice mode, you have a coach mode. uh like all of these instructions I put into this project and I gave it all the a mini version because you can't load all those 15 files in that depth to cloud pro at least when I tried this I was not able to load that that detail but I could load a bunch of these things so and it created me this copilot and let's test it and we'll probably end right after testing this so let's do medium uh these are a bunch of tests I wanted to make sure that I did before I shared with you guys. So medium level uh let's try meta because meta so the another thing I didn't give it the prep files of door dash here guys I gave it the prep file of overall analytics because I if I want the copilot to work for door dash specifically I would give door dash specific things and that would make this even more uh like specific for the interviewer targeting But let's this is like a generic product case interview uh copilot medium level diagnostics um you know investigation analysis and oops and let me call simulate on this.

42:35 Simulate is one of the modes. This is my favorite mode. It basically literally I hope this works. I think it will but let's see. So this basically simulates the actual interview for you. Uh let's say claw code basically acts as your uh interviewer and it also acts as a candidate because look at this. So this is the interviewer. Hey, thanks for joining today. I'm Priya, senior data scientist at Facebook app at Meta. Today we'll walk through the broadcast case together. Think of it as you know so and so. Let's and all we need to do is continue and it basically gives the scenario again.

43:15 So I made it in chunks because with uh in in these days we are so inundated with information and text it's so hard to process. I thought this will basically u make it less overwhelming and we'll see it chunk by chunk which is the reality right. This is how it happens in real interviews. So let's do continue again. Uh so they gave the test and look at this now Claude Core Claude is acting as a candidate. So thanks Priya that's a great problem to dig into before I jump in. I'd love to ask a couple of qualifying questions. So this is what I was asking like I was telling you guys all along that hey ask clarifying questions. Probably most of you are going to ask as you answer but look at this. So this is how it simulates the whole interview.

44:05 But let's say because we are almost out of time, I'll jump out of this and go to the copilot again and I'll show you a coach mode uh of the same thing. Coach meta rustics investigation analysis. Yes. So what does a coach do? You could also have a practice mode where you talk to it and it it basically acts like an interviewer for you, right? So this is the coach mode that it looks at and here's your question types recommendation template. It basically tells you what are the qualifying questions to ask, what are the frameworks to follow. It helps basically become a coach for you. I think we're out of time to share other types of uh like you know uh things as well. Uh but yeah, there are two other modes, practice mode and a full answer mode where it gives you the entire thing. Here it's acting like a coach. It's not giving you the full interview, but if you don't have time and you want the entire interview, it could spit out the entire interview for you too. Okay, cool. I'll stop sharing. I went over I think we have 11 minutes for questions.

45:26 >> All right. Uh ton of questions. So I'll go with the one that I have. Scroll all the way up here. Uh so Vishnu asks, I usually use perplexity deep research option for it to create a very detailed document with things I need along with references etc. So trying to see how this is unique given the non-deterministic nature most of the times. uh perplexity does pick up generic blog. So basically what's the difference between doing a perplexity deep research versus the uh multi- aent uh orchestration you're you're you're you were demoing.

46:01 >> So the the difference basically is um so what I covered yes publicity is pretty good. I tried it uh it's been a while I tried it being honest. I've been on cloud code these days. uh but uh what you saw a part of it is the deep research in the publicity covers but there's a big part of it that it uses the current uh rich database that we have from our coursework like how do you build metric trees how do you build all of that that's I I'm I think uh I mean we should try the newer perplexity but I'm not sure if out of the box could give us that today because we put our lot of knowledge about this analytics workflows and it has all the information in it. It gives I think much richer uh documents to prep for and the the the practice that you could do with the mock interview for pilot.

46:57 >> I think and then just the other thing to add is like just pretty much anything with cloud code that I really like it for is just that you can make it hyper bespoke to whatever you're working on, whoever you are. give it lots of information about yourself, what you care about, and then it's like really repeatable once you build all those workflows out. So, this has nothing to do with interviews, but like my wife used to use Perplexity to do a lot of research around ceramics, and I finally got her to do like use clawed code the other day to do like, oh, will my glaze get all screwed up if I burn at this heat? I don't know what all these things are, but she's like has now these repeatable functions where she like, oh, actually this is way better than perplexity because I can get go away faster. But I I truly haven't used perplexity for like a few months now.

47:44 >> Awesome. >> Cool. Uh, another really good question and I'll chime in after you do as well since I have thoughts. Uh, Nitton asked, uh, just curious, do you find yourself skimming these files more more often than not? And then the clarifying question on that is sometimes there's a lot of text so I wonder if anyone else ends up losing interest and starts to skim the deep dive reports. That's a good question. I mean this is something that I think I was telling you right when I was doing the interview mode in the in the copilot I tried to make it as less text as possible like because I'm overwhelmed with text right now. it's just too much to deal with. So I see where the questions coming from.

48:32 Uh but to me um I do it in chunks and that's the reason I shared with you the claw code workflow, right? You could literally go to claw code and talk to it uh and tell me hey give me the metric tree for this and you you read that you study that and then go to the next one. So the the the the research that it generates is going to sit in your repository and you're going to get you can work with cloud code to get chunks of it in chunks which is less overwhelming for you. Uh and when you have the energy and focus you could obviously go and do it the entire thing yourself uh like read through it but you could basically customize how much text you want out of all these documents. I also really like um using like asky diagrams or telling it to make like HTML diagrams and stuff if you're like a visual learner just to like break up any workflows. Like I've learned that anything in life is basically some sort of a workflow diagram and Claude is really good at translating that into something interpretable.

49:37 >> Yeah. And one thing I would say too is u find a length that you're comfortable reading and just have it sort of like get to that length. It meaning like AI help you get to that length. Um but don't skip the part of actually reading the text and like be um very critical of what you're reading. Uh the thing that probably some people have have heard me talk about is that there's been a research report about uh brain atrophies with the use of AI. The subtext there is um uh if for people who don't think and just rely on AI directly as a oneshot thing or like whatever output it gives them they just use uh yes atrophy happens for people who are critical and who is using it as as like a partner as an interactive uh peer then they actually become even sharper um so there's that very important d component so don't be lazy about it I guess that's the thing uh Find find the medium that really works for you and be extremely critical.

50:42 Okay, we've got another question from Matali. Will we be learning to build this in any of the boot camps? >> So, um the core concepts of how I did this is how you build agent skills and how we work with cloud code and the best practices of working with cloud code. We do cover that as part of our boot camp that we have like we handle and we do have a version of this uh being shared in the a analytics for builders as well.

51:12 Sean, do you want to take the analytics for builders part of the cloud code because I know you're covering that. Uh and how much of it would be the right for the the boot camp one? >> Oh yeah. I mean, basically week two of our builder course um teaches a lot of like the concepts of how this Gent system works. Um actually building starting to build stuff is around our boot camp and we're going to be opening up a more advanced boot camp as well to go even deeper into building. So if you're just interested in building cloud of code like maybe the boot camp but it's we're not building that boot camp is not about building like a interview agent workflow per se uh but any of those learnings from there can be generalized to build a like we could do that but we have breakout time and that where if you wanted to spend it on that and you wanted to talk to us about doing it then for sure you could >> and you get these out of the box if you join the AI days for builders. So I mean it's pretty cool if you want to go and build them yourself. Um uh if you know how to build skills and agents, it's not super complicated. I would say all it needs is like tons of context and tons of information and some creative thought that I know most of you all uh already are probably great at.

52:27 >> Um >> I don't know if we have any boot any of our boot camp attendees. >> Yeah. >> But I you can drop in the chat if if you think about that now. >> Yeah. I'm not going to call call them out, but I did see a couple and they did leave a review on our page as well. Okay. Uh let's see. So, next question. In boot camp, this is from Ashish Sappra. In boot camp, do you go through a standard project or can I run the project I want and you would help get it running >> in the cloud code boot camp? So we basically so uh we currently have a a boot camp. We don't have the super advanced one yet. We we will have the advanced one soon. But in the one that uh we share you basically start building agents and skills yourself and work with uh like the we have a lot in the existing free repo if you want to try it out and we'll share a advanced version of the repo called AI analyst plus that has a lot more agents and a lot more skills that you could use to probably you know build like take it and use it in your projects and we would definitely help you get started there.

53:51 Yeah, just honestly perusing that our open source repo is a really good way to learn. >> Looks like Michelle just shared uh from our earlier course about how good the course was. Thank you so much, Michelle, for the shout out. >> Yeah, >> Michelle was in cohort one. That was really fun. >> Yeah. >> Uh did I miss any question? I think we answered them all. >> Did you guys get the question about uh the privacy side of cloud code?

54:22 >> I think someone just messaged if I'm looking at the recent message until you bring the old messages. Someone just asked about I want to get better at experimentation. Would that be covered from scratch? >> We have Yeah. >> Week four. Yeah, week four is all about experimentation. So, so just to teach tell you a little about our fiveweek course in our boot camp course just to tell I know already did this but boot camp is two days and it's pretty much just like here's how you build an agentic system in cloud code um for like folks who haven't tried that before. Our our um analytics agentic system our fiveweek course is more about like analytical thinking from end to end. How do you whether you're a product manager, a data analyst, a data scientist or engineer designer, like how do you go from like framing questions to building metrics to um doing experimentation and causal inference and root cause analysis and uh segmentation to presenting all of those and getting stakeholder buying and sharing it out. We teach all those concepts like a classic kind of data like product to data class would, but it's all in the context of claude code.

55:37 So like week four is all about like experimentation. So someone who's never ran an AB test can go in there for week four, learn about how to run AB tests and learn a few causal inference me uh methods, difference and difference regression and uh uh propensity score matching. But we then teach you like here is how you can basically take all the like human judgment and analytical thinking part and then execute it technically with cloud code. So, it's kind of like we're trying to balance like the we want to give you extreme agency and the kind of like technical capability to do this, but still have all like the human judgment part in there as well, so you actually know what what's going on, if that makes sense.

56:23 But yeah, week four will be the experimentation one. I'll maybe I'll try and share some stuff out around that in the next few weeks, but it's that one's pretty cool. >> Yeah, there a couple questions. I think the question again is you can't use AI interviews. What if it's for a different company? Like I said, all these are prep material. They help you basically prep for any interview for any role and all you do is give it those uh like you know that input like which is interview that you're going for, what company it is and you prep for that and you attend the interview. You don't have to you wouldn't be using AI in the interview guys. This is for prepping. So I probably uh not sure if I understood the qu question.

57:06 >> Another question. What does your course offer beyond the anthropic claude offers for free in their learning academy? I haven't taken the learning academy from anthropic. I'm going to guess that's a self-paced thing. I don't know if it's like so this is a cohort-based course. We start and end at the same time. uh you have your async material that is like up-to-date recordings you watch throughout the week and then also throughout the week every Monday we meet as a class in the morning us three instructors and then whoever else is in the course then we also have two uh additional office hours throughout the week that are time zone dependent where we answer all questions and work through any concepts that people are having trouble with in terms of async content or if you just want to ask any rando questions. So, you have like the uh live piece of it with the instructors who are going to work specifically with like if you have questions specific to like what you're doing in your day job, like I'm happy to talk through that with you. Uh and then honestly, the the awesome thing about the boot camp was just like all the other learners and builders in there was like an awesome community. Like I I'm totally bought into this. Like we we can like learn fast like on our own doing all this stuff, especially with Cloud Code, but we like go way further together as we can learn from each other and kind of leaprog off of all each other's kind of input. So that's that's what we're trying to really build with the the cohorts and the community that's maybe not in that the learning academy through club. I I haven't done that so I don't know.

58:42 I think it's pretty good the learning academy on um like anthropic but they do talk about the generic stuff. So what we are focusing on in our boot camp and and the analytics for builders is the analytic specific use cases and one like u like I I love it myself uh uh the repo that you get if you join these courses. It's going to be a rich repo of all this information that we collected over the years and that we teach in the course.

59:11 All of that is divided into skills and agents that you could take and probably use it on any project uh at your company that you know uh you're going to like in your workflows. Oh, I'm uh I need to jump. I just looked at the time. Uh it was great. Hi and Sean if you >> I can stick around and answer questions if there's some more hang hanging. Okay. >> Thanks so much everyone. Uh I'll reach out with the free uh resources but uh Sean and I would uh take next questions.

59:39 Bye. Yeah, and we have a free Slack community. If you haven't joined, um the link is in chat. I'll paste again. Join us there. Ask whatever questions you have. We we can keep answering them there as well. >> Yeah, we can answer there too. Um do we have other I'm trying to scroll up to see what else is going on here. Uh there so there was a question around like data privacy. Yeah, >> I don't know whoever whoever uh asked that question is still here, but um so yeah, it's basically like for your company, you're going to treat, you know, quad code like any other kind of uh software AI product, right? like you're going to have to go through like procurement uh legal uh security um it to negotiate those contracts and your organization will have to make up some rules around what you can and cannot put through cloud code. So, like maybe it's totally cool to have like if you have a product to have put like just like clicks and usage data in it or something like that or like meeting transcripts, but you probably don't want to have like the actual customer data that's privy to them. Um, under some contracts, there's other contracts that's like zero data retention where um which like a lot of companies that are building AI product, that's how they do it. They negotiate these ZDR contracts. uh then you would be able to leverage it for uh for customer data. But that's like that's really more of a discussion with like yeah your your legal and security data teams.

61:24 Do we monetize this product? No, we So we don't Yeah. Okay. So I have weird philosophy with code where like I just like I don't know, man. I don't think products are gonna like I don't I just don't understand how people can monetize products when it's so easy to build products now. So like for us we just try to uh give as much as stuff away open source and then we kind of like have some stuff that we build that is like we haven't updated the new version of the AI analyst yet right we're kind of like developing that but like people in our course have access to that and then eventually we'll develop some more and then we'll like push that version out to open source but our our goal is less about like oh let's build a product that people can use and more about like dang like I think the whole industry in the data science profession is drastically changing and we want to basically like empower people with as much agency as they can as possible by leveraging these tools. Um so they don't kind of like I don't know wake up one day and be like oh shoot like I I got to get on it now.

62:35 Um, but uh, yeah, so we're more kind of in that kind of like teaching realm of stuff, but yeah, I don't know. I mean, like someone else can make this product way better than I can anyways. Like you you guys can make this if I give you this today, you'll just use it tomorrow and make it way better than I can. So, it's like it's pointless for me to try and sell anything. I think if you join us pretty pretty pretty confidently that uh if you actually keep what we teach and then keep iterating on it, you'll you'll build something way better. Well, yeah. We had a dude, uh, Laurian who was in our class and he was like, "Oh, hey, he's like in finance and he was like, um, hey, this is all like in Python and stuff, so I need to see I need to see this in like Excel or Google Sheets, and then I want to be able to trace all the formulas back there to validate the numbers." And I was like, "Dang, that's such a good idea." Like, okay, I'll try and work on that. And then he just like after the boot camp just like emailed me two days later and he was like I did it like can you check it out like where did you go from here? I'm like yeah man like you did it way better than I ever could cuz you work in finance and you know exactly what that means. Like I would just be guessing at like what a finance person needs. That's the other part of this. It's like the best person to build the product that you need is you because you know your job and your workflow and everything that's in your mind better than anyone else. Like I can't build a a product for a product manager. I can't build a product for a data engineer. I can build one for a data scientist pretty good. I can't really build like like my wife will build any product that she needs better than I can. Like even though I know her better than anyone else in the world, like she still knows herself better. So that's like the other kind of mentality around like product building and who builds what.

64:20 Cool. Maybe we'll take one last question and then we'll we'll wrap. Uh let's see. Vivc asked, "When AI is acting as an interviewer, how do you prevent it from being too agreeable? What prompting patterns actually make a push back and probe the way a real interviewer would?" >> I wish Robio was here still. I haven't been doing any of the interviews. >> Oh, I I can I can answer. >> Okay, you can answer it. >> Yeah. Um, actually, uh, uh, VC, you can actually paste this exact prompt to Claude and then it will tell you how you can ask it to be different. Try it out.

64:59 >> Yeah, I will say cloud code is far less sophantic than say like chat GPT or any of the AI stuff like all the all the chat stuff is like they have metrics that their companies are built on that are engagement metrics. So if you are working if you're using chat GPT the team that's the product team that's building chat GPT they go in and they experiment with everything they look at their metrics every day and they have metrics that's like daily active users session times blah blah blah and like if chat GPS is like mean to you you're going to drop off so that's probably why they're making it so sick ofantic like I'm just guessing but like for claude coat because on average people want to hear that they're good like right like my grandma is going to like, "Hey, look what GBT said to me." Like says, "I'm the smartest person in the world. I'm going to keep using this." Um, but like Claude Code, like if you're trying to use it for like real productivity and you want it to be hard on you, you that's that's like the magic of the whole like more like personalizic system, you can force it to do all that pretty easily.

66:04 Okay, we should probably drop I don't know if there's any other ones in here. Hi. >> Or people can drop it in Slack. Yeah, join us in the Slack community. Maybe I'll answer this last one. Shravia showed a lot of metrics. Businesses door dash docs generated for prep. Do you know how much does it hallucinate? Uh, I didn't quite catch uh the full multi- aent system, but I believe there was a step at the end to kind of like do be a fact checker. Uh so you can always dig those uh sort of like steps in independent of the generation piece to uh to sort of like control quality.

66:45 Cool. So we'll we'll send out an email after this with um we'll sync up with Stravia and get a bunch of the like uh stuff that she showed today shared out to everyone here because I know there's a bunch of people asking about that. And then um yeah, we'll share out those two discount codes to the fiveweek course and the boot camp. So the fiveweek course starts on Monday. Um and yeah, boot camp starts uh in May. Yeah, we'll we'll email everyone here that stuff um within the next couple hours and then recording we'll share out uh I think tomorrow probably or later today. Just have to wait for Zoom to get this all set up.

67:23 Cool. >> All right, thanks everyone. See you.