My first interview with Amazon, the first question was what are the assumptions of linear regression? And I fumbled bad. What are some ways ML role aspirants can leverage AI to turbocharge their chances of converting those into I always ask this question to other folks as well. How do you handle the class imbalance problem and 99% of the time people come with this single answer called smart? I mean smart is one method but if I start asking what else you can do what are the challenges of this smart and then people struggle. I'm Naman Pande. This is the Ready Set Do podcast.

Pande. This is the Ready Set Do podcast. And in this episode, my guests are Nirmal Buddha Toki and Karun Tankan. Nirmal is a senior data and applied scientist at Microsoft. And Karun is a senior data scientist at Walmart. Nurmal and Karun take us through the incredible life-changing job landing advice that the two have shared with thousands of mentees around the application process, what to expect, and how to prepare for machine learning interviews. Also, they've teamed up to write the book Decoding Machine Learning Interviews, which is just an incredible one-stop shop for 100 expertly curated machine learning questions and of course their answers. answers. These are the most common fundamental topics that are asked in interviews based on questions that are posted out there publicly and our own conversations

there publicly and our own conversations with our mentees. roughly what percentage of your knowledge gained through projects will be outside of what somebody might ask you in an interview. But when the interview comes, you need to have that optimized resource that will make it happen in those short period of time without overloading you. In line with our theme of learning from somebody that is just a few steps ahead.

somebody that is just a few steps ahead. My goal with this episode is to help you absolutely demolish your next machine learning interview. So Job right AI I hearing a lot of good reviews about them to automate your resume and recommending the right job. Subscribe on YouTube and Spotify for weekly episodes every Wednesday and daily byite-sized clips from those episodes on YouTube and Instagram. And now without any further ado, here's Niml and Karun. Welcome to the only podcast in the world featuring stories of high agency individuals who are just a few steps ahead of us. Nirmal and Karun, welcome.

Nirmal and Karun, welcome. Hey. Hey, Nan. Nice to meet you, man. Hey, guys. Thank you. So excited to decode ML interviews with both of you. I know you have a book coming out which is, you know, super exciting. But for the purposes of this conversation, I want to start off with what can somebody expect when it comes to the structure of an machine learning interview. So, we'll just say somebody is a master student, right? Probably in the US, it can be anywhere. It really doesn't matter. So what can you walk us through the you know lay of the land when it comes to interviewing? Um what sort of uh rounds are there? What questions to expect just to kind of set the stage here a little bit before we jump in and feel free to attack whoever of you wants to go first. Feel free to

of you wants to go first. Feel free to go for it. Yeah. Yeah. Sure. I I'll take a first step and let Karun add. So I would probably first say that the the very first thing to clarify is about the role your itself. Okay. Okay. Uh in my experience I have seen data scientist or ML engineer applied scientist law of variations in the role and sometimes the title is not going to do any justice. Uh it will rather confuse you unless you go a little bit deeper on the job description. get an idea of what the role is about. Uh for the ML focused or the core ML focused data science role, I think the responsibilities usually talk about building ML models, having strong knowledge of stats and uh foundational knowledge of ML deep learning uh you know some basics of ongoing genai uh

know some basics of ongoing genai uh trained you know don't have to be too deep. So unless you are going obviously for the AI engineer roles which is focused on genai but and then if you are kind of like interviewing for that kind of roles in fact our book is targeted for that category uh we are focusing mostly on the core ML focused data science roles and for that I think if we come to structure uh in my opinion there's always I I guess the technical screening uh you know the technical screening It covers the uh fundamentals of coding which is data structure and algorithm.

which is data structure and algorithm. DSA round will be there. Uh it will not be hardcore as software engineers but you know uh it will be there and then there will be some of course ML uh based interview questions and which is I think our main focus is uh from the book uh to help out. uh there are sometimes you know uh case study type questions but it's kind of rare for the core ML uh some companies also bring a little bit of ML ops uh you know how you handle uh how you can handle pretty much after building the models some companies like to also find a really sweet uh spot right between technical and behavioral. So when they

technical and behavioral. So when they bring the behavioral they will make sure they pick a project uh from your resume or let you pick a project and then discuss about the ML aspects of the project right uh those are actually some of the main topics I have seen so uh I let Kon add you know obviously I may have missed some some uh no I think that was uh pretty much perfect it's typically I say there are like a few types of questions bsnql is sort of going to your screening uh ML basics uh deep learning basics then your resume what was whatever products are there on your resume questions around that once that's done case study maybe yes maybe no depends upon the company uh for for like ML core

upon the company uh for for like ML core roles we call them like system design rounds where you're expected to uh figure out what the ML formulation is what the metrics are uh how to deploy these models and how to monitor them that sort of um uh full uh end to- end development will be there typically for mid and upwards type of roads and then yeah like normal mentioned the tech no mobility folks uh where they're trying to make sure your culture fit. So yeah and that's step typically going to be the structure. The book focuses uh right now a bit more on the theoretical MLDDL uh stat those three pillars uh those uh theoretical foundations.

theoretical foundations. Amazing. Yeah. So that makes total sense and obviously we'll be deep diving around that specific piece right that you just mentioned uh which is the focus of the book. But um something that Niml said kind of caught my attention which was that um you said when there's like the technical scream um you said that it's not as hardcore right as usual software development rules. So I just want to kind of fully understand for our listeners I'm sure some of them will have the question because the because that's the first round it's obviously paramount to clear that right because unless you do that you you won't even get to the machine learning interviews phase. So I wanted to ask you if you had any general tips, tricks or advice around how to go about preparing for that. I

how to go about preparing for that. I know people just grind le code day and night or you know just non-stop. So is that generally the way you would recommend or do you have you seen any other ways around that? I can share my perspective and and it will be based on my experience and a lot of the conversations I have had so far with my mentees you know and whoever is kind of like getting the interviews who's looking for interview help when I counsel them or mentor them uh I get to know uh like what were the expectation from the company the role they are interviewing so uh in my perspective I think for the for the ML or data science role uh the reason I said is not very hardcore is because uh it also depends

hardcore is because uh it also depends on some companies sometimes right so and and that is why I think we draw a line right between if you're applying for ML engineering title or data science titles I see you can still work as a MLheavy role you know within data science space or you can be a particularly ML engineer which is tightly embedded with the engineering focus and engineering team and their sole sole uh I think the charter or the uh role uh in in the team will be to put the model in production uh maintain its uh service health and a lot of things that actually goes beyond just building the model right right so for that kind of role if you are uh

so for that kind of role if you are uh applying for companies any top tech companies like Google Midi you know uh the ML engineering role for them there's no excuse uh they go hardcore on the uh makes sense the technical round for the uh uh coding right so I mean if someone is doing like you said gr grind on the lead code uh that'll help for those roles but if you're talking about data science roles particularly with you know uh data scientist title or even the applied scientist title in Amazon uh I would say probably the medium level of lead code should suffice right okay and and obviously if you're applying for companies where uh a lot of the graph technology becomes useful uh you know

technology becomes useful uh you know like Google Meta they work a lot of networking effect right so social network uh they usually bring uh you know uh one layer up so it it solely depends on the uh how much core engineering engineering uh acumen your role covers and if if it covers quite a bit then it goes a little bit deeper on uh on the coding side. Yes, that's what that's super helpful. Yeah, super helpful. And then so from there naturally uh you know we we will be getting to the questions themselves but something that came to mind was um when somebody is currently a student right so say you're you're in your first semester of it can be computer science it can be

of it can be computer science it can be machine learning specific program whatever the case might be like like this is an aspirant for a machine learning role um I wanted to pick both of your brains around what are some things that these students can start doing now these can any specific courses that they should take that you would recommend highly or even in like their own free time. Maybe just any resources or whatever the case might be that would really help them build that foundation, you know, right from day one instead of having to wait until the 11th hour and then suddenly you're looking at 20our days full of cramming information which obviously nobody nobody likes to do that if if they can help it. So do you have any perspectives around that? Like what

any perspectives around that? Like what are some things students can do from day one of their you know higher degrees or higher programs? Yeah, I can take this one for day one. I would say it's a bit better. I mean if your goal is to land a job with um in as short a time as possible, day one it would be better to focus on projects. Um um I'm assuming if you're going for your higher studies, you have like some uh foundation in coding. That's fair. Yeah. So uh if if you as long as you have some foundation in coding, it's best to start with projects.

best to start with projects. Uh pick projects where uh you get to uh if you're going for core ML mod core ML roles, then pick projects where you're developing machine learning models. Start out with that. There are uh quite a few guided projects. You can just Google guided projects on YouTube. Uh you can walk through them. They won't help you land the job ro but they'll help you understand machine learning life cycle. Once you have like a few guided projects under your belt, you understand what machine learning is about. You understand uh animal formulation, metric selection and fitting and tuning a model. You understand that portion. Then you can think about uh looking at like actual real world uh problems. The best place to find that is uh in my opinion is

to find that is uh in my opinion is Kaggle. Uh like businesses actually host competitions with uh prize monies. So if you check the competitions that have finished uh and if you can just list them by decreasing or prize money, it will the prize money just indicates the business relevance. So uh you can just pick projects that have probably already completed. They have good discussion boards. You can read through the discussion boards to get some ideas and uh then start building your own models. Since you've gone through the guided projects, you probably have a good uh understanding of um at least fitting in your models and the discussion boards will give you ideas about what kind of models to try, what optimiz uh optimization techniques or um tips and tricks to maybe uh massage the data, get out good features,

massage the data, get out good features, stuff like that. So that's a good way to sort of uh build good projects on business relevant problems. So if you're in your first year, focus on projects. The interview prep I feel is like something something sadly quite a bit different from your uh day-to-day as a machine learning engineer or data scientist because it's a bit more focused on lead code, SQL theory, right? Yes, theory.

theory, right? Yes, theory. It's a lot of remembering a lot of the theory and uh making sure you communicate in a structured manner during an interview. So that prep is kind of uh quite a lot different from your day-to-day as a data scientist or day-to-day as a data science student. So I would say like when you're like uh 90 days out from okay this is the point that I need to uh start ramping up for interviews that's when we can switch to interview prep mode and that prep would be a bit different. Anything to add there NL that you Yeah, I think he covered it pretty well.

Yeah, I think he covered it pretty well. Uh for me also the projects will be uh primary focus. Yeah, because I think that will showcase their skills, right? Initially the main challenge is not having professional experience on the resume, not having enough experience and how can you advocate for your skills and I think the projects comes really really handy and like Karun mentioned um Kaggle is a really good starting point because I believe in incremental learning right you don't really have to start everything from a scratch initially uh you may feel everything overwhelming you know and then uh the discussion board even the existing notebooks they will help you to understand how people are kind of uh tackling the problems and you can definitely maybe redefine the

can definitely maybe redefine the problem statement. you don't have to solve the exact same problem uh you know from the data uh and you can come up with ideas on how to kind of like redefine the problem in a little bit different way and use the same data set and you know uh come up with your own project ideas right so uh so I think that's what I'll probably focus uh on the initial time uh you know so when you are kind of like getting into the the uh runway for the for the data science or ML roles so super helpful and then I think That naturally just segus into right the you know the really meat and bones of this discussion which is that um when you're

discussion which is that um when you're so say a student has been building these projects they're actually you know trying to learn not just kind of hit execute and be like oh I did this without actually understanding what what is going on right so we're assuming that is not what's happening so say somebody goes through the paces right obviously there'll be a lot of learning that goes around with that um I guess what I'm wondering is and I'm or you know a lot of our listeners will have this question. Um what roughly and I know this is like a hard question to randomly answer but like roughly what percentage of your knowledge gained through projects will be outside of you know what somebody might ask you in an interview right so I guess does that make sense like how niche are these

make sense like how niche are these questions that I hear that they're there they're in the book but before we actually get to them you know how niche are these and how can how far can somebody get just off of their own self-arning Yeah. So it's interesting I think thanks for bringing it up. I I think this question usually comes around to me from my lot of my mentees like they are like I can do projects right so maybe five 10 projects then I'll pick the highest uh you know the good ones like whatever they feel like to uh kind of surface on the resume.

of surface on the resume. Yeah. But then the question comes like like you mentioned h how do this projects that I have done will be relevant during the interview exactly exactly or even in my work right uh so uh I believe it will be relevant in a sense that and this is where why I uh kind of like agree with uh plus one on what K mentioned is go through discussion boards because there are edge cases right because there will be things like when the problem statement is defined for example Example, uh Netflix is putting out a uh competition for recommendation engine for example. It can go on different directions like a lot of people can attempt the same problem from multiple perspectives, right? Someone can bring the

right? Someone can bring the collaborative filtering approach, someone can bring the hybrid recommendation engine approach, someone can just rely on matrix factorization a lot of things. So now when you go through this different approaches of attempting the same problem, the advantage you get is in the interviews is like when when the interviewer is covering what are the recommendation systems right now you have that uh you know some some sticky note on your on your minds if based on what you already did as a hands-on when you do the hands-on I mean uh it it kind of uh helps you a lot to structure the answer uh you know when be answered in in the interviews and they're always looking for that you know so you usually I have

for that you know so you usually I have been you know in the both seats as interviewer interview right so as an interviewer I know when someone is just swinging you know when someone is like just trying to get it through and that's just bringing some extra theory just to you know yeah I usually know like and people all the interviewers are experienced interviewers and they usually know right agreed agreed they're not looking for I would say probably uh very very depth no but they're not only looking for surface either and this is one of the thing that we try to do on our book is when me and Kon was discussing right like a lot of the times there is a I would say uh not a misinformation but it's kind of like

a misinformation but it's kind of like widely spread information that for the ML you just cover the breadth you know and and and but but I kind of disagree you know so there has to be a little bit of depth as well right uh if not then you will He kind of put in the situation where the follow-ups will uh follow up questions or the case questions interview will bring will get you and and you were like yeah the same topics I prepared day in day out and same topic got me it happened to me so it happens to everyone so wow and yeah so go ahead go ahead

go ahead same things like when you're doing your projects I feel sometimes you can uh get a bit uh niche as well in the sense that okay when I doing my masters uh my focus was primarily on uh language models uh and most of my projects were centered around language models and I I just uh assumed that that when I was uh applying for an interview uh my resume is mostly around uh building LLMs and fine-tuning LMS I thought questions would come from that I think my first interview with Amazon uh the first question was what are the assumptions of linear regression And I I I fumbled bad. And that's when I realized, oh, I I've been focusing on like this portion for so long that when it came time for the interview

that when it came time for the interview prep, I didn't have a resource to like revise all my fundamentals. And that's one thing we also want to fix with this book that hey uh these are what we think are the most common fundamental topics that you need to know and that are uh asked in interviews based on like uh fan questions that are posted out there publicly and uh our own conversations with our mentees. These are the most fundamental topics you need to know and this is the depth to which you need to know them and that's what we're trying to cover in the book and yeah I feel like this book would have helped me and

like this book would have helped me and which is why I I was really uh I I really like the idea when we pitched it and yeah really looking forward to collaborate on it. Yeah. Yeah. So I think you set it up perfectly. Right now I'm dying to know what are some of these concepts that as you said are fundamental but clearly are also critical but like you have to know these before you go into an interview.

these before you go into an interview. So do you mind giving a couple examples around what some of these things are just you know for our viewers or listeners who might be curious about that. that. Yeah. So uh the way that we structure our book uh little different than what most of the uh you know like resources out there and of course uh I mentioned this on my session about uh uh the one that we did is we we just don't want to bring another uh piece of information to the internet or to to the people. you know there's plenty of information and I always stress the fact that it's not a shortage of information it's just like how do you identify the proper signals out there is is the is the most critical step here interesting

interesting and and and when we when I myself you know like was preparing and going through the learning stuff online the the main challenge I found was not everything consolidated in in one place in such a way that it covers all the potential topics that could And then also in such a optimized way, right? The reason there is a reason we picked 100 question actually we could have gone more and and and me and Karun had a hard time prioritizing which one to pick which one to drop because the questions are uh that we had was you know like interesting and and we wanted to add but at the same time I just don't want to create another book with 500 questions.

questions. Yeah. The goal is like if someone has got an interview today right and I have two weeks to prepare like maybe three weeks to prepare of course ML is not only the section you will focus but because the technical screen is the first getway like you said right if if you if you are doing your best for that first checkpoint then you will either do some coding and for the people who are preparing I always tell that do the coding always you do one or two questions a day right you don't have to wait for interview or you don't have to be looking for job to do the coding. The ML one can probably come in that uh you don't have to like learn daily basis but when the interview

learn daily basis but when the interview comes you need to have that optimized uh you know resource that will you know make it happen in those certain time. Yeah. Without overloading you without you know uh giving you less information or more information and that was the goal. Uh so that's how we kind of plan and structure our book just to make sure that we have 100 questions. Like the other advantage for for people looking into this book would be we tried you know like to look online like what are the most repetitive patterns right and and and what are the top tech companies are asking and some of the questions are actually tagged by company names that of course it's because not the not the question that I have seen in Microsoft or not the

have seen in Microsoft or not the question that Karun has seen is in company because we don't want to disclose any of that information but the questions that already out in the internet internet you We took that kind of question and see that if we can tackle that. First we kind of like went through a process of how do we kind of group these questions and on identify the pattern and then we created those kind of uh question and answers right. So uh there's various topics of course starting from linear regression to deep learning basics right uh we purposely excluded all this uh jai and stuff because and that was not the motive of the book right uh we will we will be probably collaborating in future to expand it uh in the next iteration or next version right so amazing yeah yeah current

yeah current yep yep uh it's mainly focused on the fundamentals starting from the basics of stats and probability, basics of machine learning, what metrics you need to know and how you compare different metrics, linear regression, various regression, KN and K means, support vector machines, deep learning basics. So B it's primarily focused on the fundamentals of ML. These are the questions that uh like um my an interview experience of mine uh an interviewer said okay let's go into a rapid fire round for the next 10 minutes. He just asked me theory question one after another to the point that he was like trying to get me to miss something. So these a lot of these questions I have had uh there similar ones in the book. So I'm had that experience with like a few companies

experience with like a few companies where there okay I just want to test your ML foundation. So I'm just going to be asking you a few questions answer as best as you can. So yeah um I feel like this going to uh help a lot of people especially with their fundamentals. What I really like about what you guys are saying is that to Nirmal's earlier point around yeah there is no lack of information here but I do think it needs a special sort of skills when it comes to kind of curating which like what information there is really actually helpful and then if somebody who is clearly you know stalwords in their respective fields and have been around for so many so many years talk to so many people as much as many people as

many people as much as many people as you guys do you really start to get a almost like a competitive advantage I feel when it comes to go study this and you know thank me later type of situation. So I I do think that's a really valuable way to go about things in my mind around what you guys said that we only have 100 but these are the 100 that you need you know right so there's no point in going after 500 different questions and then nobody ever asks you that so that part makes a lot of sense one other question I had around that was um so is it mostly core fundamentals or do do you also kind of sometimes go into a little bit more of the you know advanced concept.

the you know advanced concept. So there are some advanced concepts right I I think where we draw this line of of fundamentals versus uh advanced in my opinion is like when it comes to understanding linear regressions like Kon mentioned someone has to know the the assumptions of linear regressions we make right but then within the linear regression I mean there could be other topics where they can go a little bit deeper on residual side you know how do you analyze uh you know the the patterns in the residuals plot you know things like that uh that doesn't necessarily I mean I would probably say part of fundamentals but at the same time going to swing a little bit towards the the knowing a little bit more than just knowing the

little bit more than just knowing the assumptions right so a lot of times like people I have seen appear like okay I know R square is is a popular metrics to to measure the regression evaluation but then if we ask there is a problem in the R square and do you know what problem that is and then can you explain what is the solution or how do you tackle that and and and it's actually not too complex question in my opinion it's not in the advanced zone yet but a lot of people I I would say in my experience when I brought that you know followup or or that question after square 50% of the people miss out and then the rest 50 also sometimes not everyone kind of

also sometimes not everyone kind of answers to the fact that hey always adding the feature is not always uh you know we cannot guarantee that adding a feature is always improves the model performance. It may slightly improve because that's that's the that's how the linear regression works. It it may slightly improve but it's might be just a disguise right so and and and and how do you tackle that you know how do you penalize it so adjusted R square why it existed or people don't know that so it is basic things I would say right but it goes a little bit beyond the basics exactly exactly so our goal is to to not only kind of like revolve around the basics because

like revolve around the basics because those are covered like I said if we only ask a question about what is PCA how it is done that is covered covered by everyone but will like hey go a little bit beyond that right what are the other alternatives right uh why why what are the disadvantage of or or the flip side of you know where PCA doesn't work for example uh and this is one of the thing I covered on on our last conversation also like most of the times and even when I used to prepare I did the same thing I was like what is this concept what it does what is the advantage of this I never looked in the flip side every concept has a flip side So we're trying to bring that also in in

So we're trying to bring that also in in the book. So that's that's another advantage uh people will get. Yeah. Yeah. I think that dates with what Karan was saying around the edge cases, right? Like the edge cases are important. You have to have them covered or they will come back and bite you at some point. Of course. Um, one random gripe that I have with when I try to pick up technical books, right, of kind of this nature, something that, you know, stands in my way is just the future proofing piece of it, if you will. Like I feel like a lot of books that are written around um like technical stuff or especially around coding somehow don't always stand the test of time. Like there are some that do of course, right? But some just like

do of course, right? But some just like as in if you picked up a book on I don't know like some JavaScript framework. It's just sad because probably nobody even like uses that anymore. So so I understand that we already kind of avoid some of those problems with machine learning stuff cuz these fundamentals are just code right they will forever be around. But I did want to ask if you have specifically taken any steps to you know future proof this to the highest degree possible especially with AI moving at the speed that it is. Uh yeah does that version make sense? Is this something that you thought of at all?

something that you thought of at all? Uh so in terms of like futurep proofing the book uh I guess we came from an angle that these fundamentals are sort of how models will be improved over time. So um once you build a model at the end of the day you have to do error analysis you have to figure out where it's uh not doing well and then based on that you have to go uh and figure out why this particular model is not doing well in those regions and eventually it's going to come back to metrics uh bias variance trade-offs uh hyperparameter tuning uh regularization some of those core concepts uh it's going to eventually come back into those. So really understanding like if

those. So really understanding like if your goal is to work in core ML uh really understanding how these models are built evaluated uh sort of becomes uh necessary and since that is the focus of the book even though we didn't plan on it since that is the focus of the book it sort of makes it self sort of future proof it takes care of yeah just by the definition yeah that makes sense. Yeah, just one slight thing will be like I mean like K mentioned most of our stuff and you also uh in the in the question itself you had answer in my opinion because we we're covering core fundamentals for the most part and this concepts will not change right uh so in that sense I think the only time when

that sense I think the only time when the book will be irrelevant would be if the company decide to do that hey don't we are not going to do ML interviews that's not going to happen in next few years for sure right right that was happening the the first thing that should go away is the coding rounds and the companies still do the coding rounds right so proof so I think yeah until that happens that if if ML rounds uh are like hey we we just don't want to do it then only people don't need it but besides the point that that point I think our book should be should be usable for for uh long run you know it's not about just any versions of like you said JavaScript or that's another thing

said JavaScript or that's another thing about writing the core technical book the challenge is there how do you kind of like follow up with the uh with the different versions that you taught in the book yeah I've come across se several books where uh you know the things don't work so so that's not ideal right then like that defeats the entire purpose of but if you know like a technical book people usually keep a repo GitHub repo for at at least the coding side and they keep updating, right? And and and it's actually should be the right approach when you release a book. If you have like coding involved or something that you think will change in future, uh you change the gate rates.

uh you change the gate rates. That is not even something I knew actually. That is that is fascinating. Yeah. Uh Kar something. Uh no no uh I I sort of agree. Yeah. Yeah. Yeah. Um and then so are there a bunch of like code examples in this book as well or does it usually deal more with the theory side of things? we don't have much tooling uh and the like the goal was not to cover the yeah the data structure or the coding part right uh there are some problems that are actual problems solved you know uh in the math and stat side probability around that area and of course the the concepts will also remain right the equation won't change so it is future proof in my opinion opinion no absolutely yeah and it definitely

no absolutely yeah and it definitely sounds like it right I for I for um I I can't even wait to read this and I'm not even from this field you know so I don't plan on interviewing for any ML roles anytime soon but from what you've laid out it and just because I'm curious as a person and I do have some background with you know writing code and such even though I don't do that anymore yeah it is something that fascinates me so I'm just happy that you know I can pick pick it up without any issues and you know hopefully that'll that will work out. Um yeah so I make sure you get a copy. Thank you. Appreciate that.

Thank you. Appreciate that. And then from there I did also want to talk about um obviously you you would have to you know go back and forth with hey let's include this question let's not include that question. Can you can both of you maybe nominate like one of your We can have it be favorite slashmost underrated question whichever one you want to tackle but I would love to you know have a two question slash that are kind of closest to each of you individually if that makes sense.

individually if that makes sense. Um, for me I I guess the question that got me into this prep, the question that I fumbled the first time around, assumptions of linear regression is sort of like my favorite question. I still sort of use it as an example. Wow. Whenever I talk about like interviews. Uh, yeah. Because I sort of fumbled on it. I was like, okay, I got it. Personal connection. Exactly. You must feel so extra passionate that I will not let a single other person trip on this.

single other person trip on this. just because I have I love that right that's that that makes so much sense what about you yeah yeah one example for me would be uh and then again I mean do not quote me that the same question exactly I just in the book because my mind is kind of numb because we went through a lot of oh yeah I can imagine stuff you know so uh but yeah like Karun said I was also you know like I I picked up some of the questions where I got uh you challenge from different angle. I picked up those questions as well to include uh you know so one was basically

include uh you know so one was basically the smart thing you know like I discussed on our conversation last time too is uh and and class imbalance exist in real world right so there is no uh and I work in security where we always have class imbalance on any of the data because normal behaviors are way too many than uh you know uh kind of like abnormal or anomaly behaviors. So uh we we bring I always ask this question to other folks as well because uh kind of stick with me and like hey how do you handle the class imbalance problem and 99% of the time people come with a single answer called smart right so we do smart and um I mean smart is one method I don't disagree but and then if

method I don't disagree but and then if I start asking a little bit of you know uh kind of different way like hey besides the smart what else you can do right right what are the challenges of the smart uh and then people struggle uh and and I used to be on the same seat. So you know like I I usually know like if people prepare just surfacely then they only remember a few things that exist out there like hey uh do smart and you you're good to go you know but then if you understand hey you know there what makes get you know uh there's some own challenges drawbacks right so if there are noise it's going to amplify the noise uh you know so there are lot

the noise uh you know so there are lot of things and there are actually few basic things you can do like if you if you learn regularity ation for example uh you know you can also apply you know like if the model is making errors from the minority class uh punish them more right uh it's a simple techniques and ras kind of like is around the same concept uh you can kind of like use the class words uh exactly the reverse way of ratio right and and scikitlearn this packages provides actually good parameters a lot of times people don't understand all the parameters they only look into the the the basic parameters, right? But, you know, a lot of these functions and stuff has actually, you know, more functionality, you know, so

know, more functionality, you know, so more more things you can do. Uh, so you need to explore. Wow. Yeah, that's that that's sounds super airtight and I love how you how you covered every angle there. I think obviously, you know, that's it's really important. Um for our last segment here, I'm very curious to explore how um obviously with the rise of AI, you know, it's kind of hard to go anywhere without hearing that term at this point as of recording this in um July of 2025. But what are some ways according to you guys that students or any really ML role aspirants can leverage AI maybe in conjunction in some way or form with the book maybe not doesn't have to be um to you know really turbocharge their chances of converting those interviews

chances of converting those interviews uh there are actually already a couple of good sites that have like LLMs integrated into them. I think Google even uh launched their own interview prep site for behavioral rounds or you can just talk uh talk to a uh agent and it'll um like record your answer and evaluate you on that. There is another u one of our sort of like u creative friends uh Don Chu has a website called interview master. So it's mostly around SQL and recently around Python. So whenever you're coding if you get stuck you can ask for a hint and and an LLM model analyzes your current code and uh provides you suggestions based on that and uh yeah it's good in the sense

that and uh yeah it's good in the sense that u typically like lead code for instance you solve a program solve a question and then that's it and you go to the next question typically in interviews it's not like that there will be follow-up questions uh but you need like a human to understand what you have written right now ask those follow-up questions So uh a few platforms interview master is one example that it's able to analyze your code and uh like uh uh ask follow good follow-up questions. So there are like uh a few platforms out there. If you want to go just uh core like chat GPT, I've shared like a few prompts on my uh LinkedIn where like if you use this prompt, you'll get like maybe like a out of like it'll give you like 10 questions and probably like six

like 10 questions and probably like six or seven are pretty decent questions and you can just use those questions to I mean you can ask RGB those questions and figure out those concepts. So correct correct like the base concepts like linear regression, lover regression, those models, you just ask RGBT, hey give me like uh there's a particular prompt that uh I I recommend people use and if you do that it'll give you like 10 questions on that topic and you can just try and answer those questions yourself. So that's another way of doing prep. Yeah, nice. And probably you can use that in projects as well, right? Cuz you can just have Claude or whatever be your copilot when you're building out any project. I know Nvidia has a site as

project. I know Nvidia has a site as well where they let you tweak projects just directly on the cloud itself. Um so yeah I I think in terms of the code piece obviously there's you know many use cases there but yeah I didn't know about uh both those tools that you mentioned that the one that Google has and the sept one that John has um set up. So both of those are yeah sound like amazing resources for our listeners. Um, normal. normal. Yeah, I think uh I would probably use the fundamental again here, right? So like K mentioned just like prepare some good prompts and and interact with chat GPD. Save that session, you know, and and you can continue chat on that session. Uh sometimes you have to a

session. Uh sometimes you have to a little bit customize based on the role you're interviewing, right? Don't blindly just think that any AI tool is out there that's going to help me out and there is not actually. Yeah, they are just a structured approach uh of customizing to your personalization or to the right job. So what I would do is probably create spend a little bit time on crafting really good prompt and understand hey this is the role I'm interviewing you know like based on the recruiter conversation take some notes and and put that notes you know on the on the prompt like this is what the recruiter has mentioned the exact job ID go to that job ID grab some content like the useful one the responsibilities you

the useful one the responsibilities you know whatever is listed in the role add there you know this is the role I'm interviewing and I'm looking for kind of like a structured approach of preparing So and you'll have that back and forth conversation uh and uh ask few sample questions you know and and what could be the uh age cases around this question. So you have to be more creative right. Exactly. Exactly. So I think the more creative you are the the more uh you know the in-depth uh reasoning you can get from these models.

reasoning you can get from these models. These models are actually really good. And when it comes to exact tools, I guess there are uh from the hiring aspect, I think from the resume prep to to kind of interview prep, there are many out there, right? So, uh job right, AI, I I hearing a lot of good reviews about them uh especially from the beginning itself to kind of automate your resume uh and and mapping and recommending the right job.

recommending the right job. Nice. Uh so from the Google I think uh notebook LLM I hear that if you if you build a project and if you have a notebook if you created a notebook then or if you even want to convert anyone else notebook into uh some sort of listening podcast or you know like yeah content like some and it's good like if you're just uh doing nothing and you know just walking around or or just driving then and listen to it you know and and yeah I agree it's I use it all the time actually and especially the mind map feature that it has. I found I find that to be so helpful. I'm a very visual person, so it really helps me kind of see the network kind of come up and have all of the

kind of come up and have all of the various nodes interact with each other and go off of each other. So yeah, I think that's a great call out. And maybe one other thing that you know in my mind makes sense is and this is just something I do. So I really like using the like voice version of Chad GP, right? So I Right. So I feel like all like sometimes typing stuff is not the most efficient way. So um sometimes I'll just have it be like hey you know I want I'm trying to learn um really robust fundamentals around economics 101 like it can be anything in the world and I tell it and I prompt it to ask me

tell it and I prompt it to ask me questions. So it's not like a mono drone, you know, speaking in my ear cuz then I just tune out. But I tell it that assume normal intelligence, assume normal worldview and then quiz me around learning to or kind of teaching myself just like key concepts around economics and maybe that is also something that students could do here where they can insert the job description, the role title, whatever the case might be and kind of you know just do a literal interview with using the like voice feature probably. And one thing I would recommend is do mock interviews. Mock interviews help a lot. I mean this is one of the underrated thing that people I mean the one of the reason is most of the mocks

one of the reason is most of the mocks are usually expensive as compared to any other packages out there like someone has a resume review session versus the mock interview it's a little expensive that's because they are actually uh giving you the real simulation of how interview happens right so uh I do a lot of mock interviews for the folks you know and then it helps out because in in the past I didn't do it much right so uh and then I kind of see the value behind it actually uh you know it just gives you a little bit extra confidence sometimes at the same time I think you know it's it helps you right uh tremendously so yeah I love that do you so and the the

yeah I love that do you so and the the way to do that is normally hopefully find somebody that's you know really top of their field and just politely request them to do that There are experts right like the talk is the popular one and myself and kun we both are there you know like find our services there uh you know and I've been doing mocks and other mentoring sessions including resume reviews and all right so got and there are various websites also that actually has the uh mentees or mentors who uh will do the mocks right so got it uh you just have to look out uh go through their uh kind of like review process is you know make sure that you are getting the uh right deal for the right money you know of course because if you're spending uh but it will help in my opinion for sure

in my opinion for sure no totally I I mean I 100% agree I feel like even if you remove the ML part of the equation any interview right you'll be you'll be it'll be hard to find any interview where somebody's chances wouldn't improve if they were actually not just with their friends right but actual strangers are helping them you know prepare for that interview. That's Yeah, that's such a such a good call out. And so so in my mind, right, I feel like both the both of you have made it abundantly clear who the book is for, why somebody should get it, what the point is, and what they get out of it.

point is, and what they get out of it. So all of that is completely clear. I do want to open the floor for each of you to kind of address or talk to people that are still a little bit on the fence around getting this. And is there and I'm curious if there's anything that you want to say to these people if at all.

want to say to these people if at all. You you can also just be like, "Nope, no worries." But uh yeah, I mean, if you're on the fence about the book, uh yeah, it's it's sort of like the most concise way of getting your fundamentals refreshed. Yeah. uh I I I usually used to recommend what I call uh like two core books which is ISLP introduction to static learning in Python. It's very popular book and deep learning by Ian Goodfellow. He's sort of the person who started the entire generative AI wave. Wow. Wow. Uh most of the concepts that I learned from these two books uh I have uh used

from these two books uh I have uh used to answer the questions in this book. So now this is going to be my go-to book for for uh recommending for interview prep and uh yeah. Yeah. I I think if you're uh on the fence uh it's going to be the most concise way to refresh your fundamentals. fundamentals. Yeah. Just like your cheat sheet, right?

Yeah. Just like your cheat sheet, right? But except it's the only cheat sheet that you will ever need because it had it actually has everything. Yeah. I love that. That definitely convinced a few people right there. Yeah. So, so for me I think uh I would probably ask what else you are looking for uh in a book if you're preparing for interview where two of the like-minded folks have come together uh they bring all the years of experience they have gone through their own interview and then you know like taking and giving interviews so that actually brings a depth of knowledge and we are trying to help to succeed right?

we are trying to help to succeed right? Uh the candidates uh and I already said before uh don't be in the trap of too much information out there. Right? So I will tell people to do your due diligence but we are not going to price it so high that you will regret. Uh for the buck you spend this book will have everything for you. So amazing. amazing. Yeah. And in case you want a sample we have a website. We'll get a sample five questions. If you're still not uh sure, both Nurmal and I have been posting interview questions for well over a year. year. Yeah, there you'll see sample questions there as well. And obviously I'll be linking it, you know, the site where you have that in the description for anybody to check out

the description for anybody to check out for sure. But yeah, um thanks so much guys. Thank you so much for taking the time. Um, I'm like I said, this probably is not much of an endorsement, but I'm not even from this domain and I'm actually curious to, you know, like flip through this and, you know, just kind of brush up my good old concepts with like cobwebs hanging around cuz it's been so long. But I think it'll be interesting to get back uh into things. So, really appreciate you both taking the time here today and all the very very best for the launch. And also finally thank you from and like on the behalf of all of the like hundreds of people combined that you help uh and have been helping through the years. So yeah really really appreciate you all.

appreciate you all. Yeah thanks. Thanks I mean thanks for inviting us. Yeah thanks for having us. It was our pleasure uh and uh you know uh we'll connect offline and it was nice meeting you. Likewise. Yeah. Cheer. That brings us to the end of that episode with Nurmal and Karun. My mind was absolutely blown by just how much content they managed to put in that book. If you would like to buy that, the link is in the description. If you would like to support me, the easiest way to do that is by subscribing on YouTube and leaving me up to a fivestar rating on Spotify or any of your favorite podcast apps. Something that goes a really long way for me is if you comment on my videos and just leave feedback, whatever the feedback might be. Finally, if you

the feedback might be. Finally, if you liked it, please do share this with a friend or family or really whoever would care to listen. Catch you all in the next one. New episodes every Wednesday.