Hi, Im Ayush Agarwal , from IIT BHU ECE (Electronics ) 24
This repo is meant to guide juniors for data science placements, specially Infoedge DS .
Do star the repo and tell me if u find it useful (honestly just tell) , took me some decent time to write all this :) I would be happy to talk to you
I got into Infoedge as Data Scientist , had 12 shortlists and 2 off campus opportuinities
Shortlists – Infoedge DS , Sprinklr DS, FPL Technologies DS, JM Financials DS (DS is data science) , AspectRatio Analyst (DA is data analyst)(waitlist) , ZS consulting, Samsung Noida SDE , Micron Core ECE , Qualcomm core ECE , Flipkart AMBA (Assistant Manager business development profile) (managerial role) , ICICI management trainee (managerial role) , Sprinklr Product Analyst (Product Management role)
Off campus – Google Hardware (reached till round 3(final round)) , Tata Elxsi (Mechatronics/Automation engineer)(reached till final round)
So I am multifielder , meaning I had experience across various domains and profiles ( I won't deny that I am flexing here XD)
My linkedin - https://www.linkedin.com/in/ayush-agarwal-261041215/
- Data Scientist – Passionate about building AI, crunching data, making ML models, plotting things, magic like things with pictures text voice tabular data and much more, then data science is for you. Ive explnaed in extreme depth everything about this profile , just scroll below. Ur target companies are – navi , hilabs , infoedge , sprinklr , fpl , publicis sapient . should know dsa too, can also aim for analyst as backup .
- Data Analyst – excel , sql , graphs , data interpretation needed . companies – axxela , axtria , citi , futures first etc (I don’t remember many but there are tons) . Case studies and guesstimates asked. Has more jobs but lesser pay as compared to data science.
- Product Management – also comes under names like apm (associate product manager) , product analyst (pa) . This is a business role that comes after MBA usually . Few ask you to make decks like Flipkart APM and Hilabs APM (hiring numbers very less in these ) . Other few ask you aptitude , product sense etc . more Companies – sprinklr pa , meesho apm , medianet pa etc . In interviews there will be emphasis on product design , RCA (root cause analysis), product cases etc
- Managerial role - This is a business role that comes after MBA usually. Honestly , very small profile , nobody even told me this profile exists , and there is no hard skill they test . you should have good English , good communication skills , be able to cook up and present stories well and have business acumen . Even Im not sure what they want , but about 10% of iim prepare for this role . people skill is a key here . companies – flipkart am ba , icici gm
- Consulting - This is a business role that comes after MBA usually, rather I would say MBA colleges ki “coding” . Black coats, good exposure , foreign travel and long work hours . companies – exl , zs . should prepare for guesstimates cases . should remember frameworks for all types of cases like RCA (root cause analysis) , market entry , profitability and prepare aptitude and English well . (both exl and zs are mass recruiters do apply)
- Core ECE – I made a whole github repo last year after I got my intern in nvidia – github ayush-agarwal-0502 electronics-guidance-freshers . Verilog , digital electronics , mosfets vlsi etc companies - Nvidia , Texas Instruments, Qualcomm , Micron , signalchip , edgeq ,
- SDE- coding . coding . coding . do dsa , os , oops , dbms , computer networks (cn) (start watching gatesmashers atleast 6 months before if u want sde only). Candidate may do system design (low lld and high hld) and machine coding if they have time and are preparing specially for this . Ask a coder for tips , I didn’t have much intrest here so Ill skip this . joboverflow for questions , interviewbit for interview practice , leetcode for coding , gfg for article might be useful for you
- Quant - intrested in working 12 hours a day devising algorithms to model stock market and get money using automated trading , using heavy maths statistics ml and finance ? These are the 1cr jobs , very few exist , and ive never seen anyone passionate about hft or making their own trading bots (worldquant dosent count) . companies - graviton , plutus research , nk securities . for this study probablity , statistics , brainstellar puzzles, little bit of system design and os level implementation and boosting of algorithms . Also they ask for good cf rank i suppose . this is the profile you are looking for if you only intrested in money .
- Off campus - Intrested in some very specific domains like CyberSec, UI/UX etc ? If companies of your preferred domain don't come for hiring you can always apply off campus tho it would definitely require hardwork.
- Research - If you have a single subject and field of study you are passionate about , dont like corporate , want to settle in foreign, maybe become a professor at a uni as a side job, then this is for you . contact research committee of our college for guidance on this . Mostly would be required to crack scholarships like mitacs charpauk etc (idk much here)
Scope har jagah hai , har profile ka apna apna scope hai , cant emphasise it enough - don't blindly fo into SDE !!!! Explore , see what you like and choose your career accordingly . Even in the worst case you would still earn enough money to have a decent life as compared to any non iit (no offence meant) its just that relatively thinking about money makes people feel low , so use this freedom and do what you really like , after all you gonna do the same thing for 8-10 hours everyday in your life . Also consult more seniors according to which profile you preparing for as my explanation wont be exhaustive :)
MAS Interview experience video : https://www.youtube.com/watch?v=nWyBQhGnHaM&t=20s
Paper had 40 mcqs , all on mixed concepts , I don’t remember much of it honestly . Some questions I remember are - what false negative means , calculating variance of a continuous function over an interval, something about alpha and beta for type 1 and 2 errors, something on kurtosis, homescedasticity vs heteroscedasticity, dickey fuller test, bag of words, which is text processing - stemming lemmatization tokenization, something on what eigenvector and eigenvalues mean in PCA, types of ensemble - bagging boosting stacking, something on xavier and he initialization, difference between append and extend in python, pandas .nunique(), pandas to_csv(), DROP in SQL, Pass in Python usage, SELECT DISTINCT in SQL, ,describe(), .reshape(), slicing in python, resnets - for curing vanishing gradient problems, which are classification model errors - cross entropy log loss hinge loss, deep belif network dbn and rbm restricted boltzmann machine, random forest is bagging or boosting - bagging obvio, sum of 2 independent normals, universal approximation theorem, OvO OvA one vs one one vs all methods of multiclass, bias variance graphs with error-model complexity and error-iterations, number of hidden layers in deep nn - 1 , pandas - merge join concat, which is faster - stochastic gradient descent or batch , append in python, lstm used in - speech recognition, entropy formula, mutable data types in python
Then there were 5 rounds of interview lasting almost 5 hours
First I will give overview of the rounds then tell about the questions
First and second rounds were continuous grilling on variety of topics . They made sure to ask questions from SQL, Python, ML, DL, Statistics, Probability, Linear Algebra Third round was ML coding round on a colab notebook made by the interviewer 4th round was case study and background round 5th round was HR round
There were lots of questions, some questions I remember are –
Probability and statistics – Bayes theorem , formula for bayes theorem , assumptions behind bayes theorem (mece – mutually exculsive collectively exhaustive events) , how will you use it in real life (basically I directed this question to naïve bayes theorem), they also asked me which distribution can be used for modelling number of customers coming or something like that (the question was related too poisson so I said poisson distribution ) then they wanted to know what all poisson distribution could model (so I explained about how it can be used to model any count data, number of customers coming in shop ,number of cars passing etc , and then took them towards GLM and poisson regression , they were satisfied) ,they asked geometric distribution or something which I didn’t know then they asked me to make fn for probability that it will take n failures before first success ( I took the assumption that events are Bernoulli, then derived it on the spot , then when I came with the final answer they said yeah this is called as geometric distribution ) , then they asked what is linearity of expectation (basically E(aX+bY) = aex + bey ) , p value and alpha significance level
Linear Algebra – Rank of matrix, Null space of matrix (linear algebra is the only topic which only infoedge and no other company asks , if you want to go in infoedge prepare this also)
Deep Learning – this was the most drilled section . They asked me about vanishing gradients , exploding gradients, how to avoid it (xavier, he initialization for relu and tanh respectively I said) , weight decay , how to avoid overfitting in neural networks (dropouts) , what is learning rate , how to adjust learning rate , intuition behind how neural networks train and how backpropogation works, They asked me what to do if neural network gets stuck at local optima (you rerun it again and again and take the lowest loss fn value iteration ) For CNN side , they asked me to explain it (I explained the filters , stride , padding , 3 parts convolution activation pooling and n+2p-f/s+1 vaala formula they were content with me after that) they asked me what max pooling is , they asked whether CNN is rotationly invariant or variant (whether it can work on rotated images or not) , whether it is translationally invariant or not . They asked me what deep learning model to use for time series (then I explained them about rnn gru lstm and bptt (backpropogation through time) low memory problems of rnn , little bit of lstm gru structure and they were content with my answer)
In the 4th round when I said I knew bit of RL (reinforcement learning also) , they asked me bit on that also so I explained environment agents and reward functions and policies for learning and Markov models and DQN algorithm . He also asked me about itertools and oops in python , I explained him about oops principles in c++ and told him Ill adapt to python when I work.
Classical ML – They asked me how I can reduce overfitting in trees (pruning) , as my project has PCA and t-SNE , they asked me how PCA works (I explained SVD Singular Value Decomposition Scree plots and % explained variance and linearity and how it’s a deterministic algorithm , they were content with my answer) , then they asked about t sne (I explained it was made by geoffery hinton in 2008, successor to sne algorithm explained about how it’s a non deterministic dimensionality reduction algorithm , how it can capture non linear relations like swiss roll datasets also , how it tries to preserve neighbours in projected lower dimension and thus good for visualizing clusters, little bit on perplexity also ), then they asked me about LDA (Linear Discriminant Analysis alg or something, I explained them how it does dimensionality reduction while maximise fisher lda formula and increasing separability of clusters , how its deterministic and linear algorithm etc)
They asked me bit on Markov chains also (I guessed it was about FSM diagrams and transition probabilities and Markov chain is related to chain of events to which they replied yes that’s correct) In 4th round , they asked me what is heteroscedasticity (opposite of homoscedasticity) and how we can model data using linear regression which disobeys the homoscedasticity assumption (I answered that I would use Generalized Linear Models (GLMs) and then model the variance also as a linear combination of inputs and use this model . Although they said they were looking for some transformation to make it homoscedastic (I jokingly said standardize it) they were supportive of the GLM answer too)
They asked what SVM is (then I explained on svm , hinge loss , kernels , and how kernels project datapoints into higher dimensions to find hyperplane and linear separablity , bit explanation on support vectors and what margin is )
They also asked me a bit on multicollinearity , how to detect it and how to avoid it (I answered about VIF variable inflation factor , 1/1-R^2 , the intuition of multicollinearity , dropping columns to avoid it )
In the ML coding round , the dataset was on MBA students , we were given few numerical columns and categorical ones , and we had to predict status of placement and salary column was also given . Documentation was allowed but yeah its not that hard to remember some of the code also if you really are an ML person it just comes . They asked me to predict status of placement using other columns , I could see that Had to check missing values (there were none) then they verbally asked how would you have dealt if there were (I explained dropping , imputing mean median , imputing mode for categorical , or using xgboost imputer for filling missing values, and I wrote it in comments also so that they could keep track ) , then we had to check outliers . But Beware , they were checking presence of mind , since all of the data was meant to be percentage marks in exam , they had to be in 0 to 100 range , printing data.describe showed me that values were roughly between 30 to 90 range , so I could clearly say there were no outliers . Applying box plot or z score to remove outliers would have given an impression that you rote memorized it without looking at the data , so be careful they might use tricks like this to test you . Then had to ordinal encode the categorical columns . Then had to print value counts for each column , and say that data was imbalanced . he asked me what methods can be used to tackle imbalanced and I took a deep dive into undersampling , oversampling , near miss algorithm , smote , borderline smote , adasyn etc to which he said yeah you can change the weights also for the datapoints inside the model to which I agreed . But he said that we were not expected to balance the dataset for that 1 hr . Then we plotted few seaborn box plots , and did train test split , and trained a logistic regression model , then made the confusion matrix then wrote accuracy f1 score precision and recall .
In the 4th case study round that was hardest as I was grilled until the point I said I didn’t know . First he wanted to know my background about where all I had learnt ml from and where all had I used it so I gave him a deep life on how I used rl in robotics club , computer vision experience from inter iit , got business intrests , picked random datasets to work on online , branch fest udyam bla bla bla . This was the hardest round imo (boss level vibes) and the interviewer looked quite senior also . Then I was asked 2 case studies – If I was the sales head of Naukri dot com how will I improve the revenue . I thought a lot , then explained that revenue would come from postings of jobs by companies , so more companies means more website users and more users meant more companies so the customer segments grow symbiotically . So we can target new customer markets , we can have tie ups with colleges so that they don’t need to build their tpc portal , and students can register on our site . this would also increase our user acquisition . Imo I think he was testing bit of company knowledge too since I had to know how Naukri earns money . Then I further went to customer base expansion point , and went into how I know that France people prefer French over english , so we need to make our site multilingual and adapt it to different countries to get more customers . and the last point I made is that doing analysis of current customer bases and figuring out what our competitiors do better than us . Then 2nd case study – imagine you have all the browsing data of a customer at Naukri dot com , you have to find whether they will buy (finalize the deal) on a certain property or not ? – make features (he kept saying more more until I get stuck) I made a feature using browsing data that if someone sees most properties in 30k range and few in 20k few in 50k then they wont take 50k or 20k one . I made feature using locations and emphasized its importance by giving examples of Noida and bangalore . I made features on history of customer contacting broker , property visits arranged by them etc and the interviewer was convinced .
The 2nd 3rd and 4th rounds were literally consecutive , happened back to back . they literally told me “tum 5 minute issi meet me ruko hum next interviewer bhej rahe hai” (kaafi Synergy thi mere and company ke beech, when I came out I realized bahut logo ka atka hua tha 2nd 3rd pe hi so I was really blessed here) , so my tip here would be to stay hydrated and practice speaking continuously for 3 hours , since you might not get breaks . Then I went to give qualcomm round 1 (that interviewer was lowkey annoyed why I came so late and so wasn’t interested in me) . Then I came back to infoedge and gave round 5 .
In round 5 , it was HR round , they asked me to introduce myself . I focused on showing leadership and management skills now since this is HR round intro . then we talked a bit about guitar (I guess he also plays guitar lol) He asked me – what does infoedge do (and I explained about jeevansathi , Naukri , 99 acres , shiksha etc multicoglomerate , internet company etc) and the hr was happy . then they asked me – why infoedge (then I explained about how infoedge does end to end work , makes sure ML models are serving business purpose and helping real people , and they are good at ml , explained some stuff from what I saw in ppt (yes I watch ppts , yes it helps , yes I recommend my juniors watching atleast for their dream companies) and explained about how I was impressed with the research group they menthioned in the ppt . HR was impressed with the answer . then they explained me basic stuff like 2 offices in noida , 4 day wfo 1 day wfh etc (To which I said I would love to come to office and connect offline with seniors and learn as I am covid batch and ik ki online is not good) (Insert a brick of butter XD) round was done in 15-20 minutes and was pretty chill . Finally I was taken to the TPR room and he said – koi aur interview hai ? chill kardo usse , congratulations and that’s how it finished : )
The test had mostly English , very less aptitude , and 10 economics questions of exact same type – if at 20 dollar price you could sell 100 and at 21 you can sell 90 at what price will you have max revenue and how much is the max revenue . answer to it is that it will be a parabola – R(x) = (100-10x)(20+x) and you can find maxima easily , I remembered it from my jee time so was very easy for me .
I was confused while going to the interview since I was told I got infoedge ds in 1-1 and slot 1 results were expected like 3-4pm but didn’t come so I was panicking but decided to sit for 1-2 anyways . (slot 1 result came next morning , a little heads up would have been nice but yeah ok)
The interview started with interviewer giving interview and he was from clustering department in flipkart where he clusters customer segments and decides stratergies . I had my project – starbucks customer segmentation so I pretty much glowed up .
I gave my intro explaining how my intrests went from engineering slowly into management . then he said lets start with your project starbucks customer segmentation . (obviously) . He asked mw the story behind it , and we discussed business models where we earn money from interest on the money deposited by the customers . He was satisfied with my knowledge of finance . I was explaining k means and PCA but the interviewer was not interested in tech at all and asked me to stop on that
Then he had questions –
You would have chosen few important features out of the 39 features , what were the important features – so I explained him how pca creates a lower dimension version of actual dataset and the axes represent a mixture of the original features
What were the segments formed – I took him to my github page and was explaining him the clusters .
After getting to know the clusters , he twisted and made more questions – imagine we have 3 clusters as you have in ur project – 1 grp which is regular office goers , 1 is discount chaser , 1 is old ppl who come irrespective of offers , which one will you expand , I said that these decisions should be based on the revenue generated from each cluster and went for the 3rd cluster using my intuition and he agreed to my answer saying that he thinks the same . Then he wanted me to know which cluster to expand next – then I told him office goers – and took the discussion towards cac and word of mouth marketing strategy , explained him how office goers carrying our starbucks cups would literally be a free edvertisments for us (one of my strengths is marketing , I have good intuition on marketing strategies so I took the discussion to that side) We had some discussion on marketing strategies and he agreed with my answers . He also explained me that in flipkart , customer acquisition cost is pronounced as kaaq . I told him that we can test a control system and reduce the number of discount offers given to regular customers and although I was shy about saying negative things about customer , he reinforced me by saying that in real life its not just about the comfort of customers , we have to optimize our monetary resources in such a way that we can serve larger customer base more efficiently , and agreed to the marketing strategy I designed . Then he wanted me to explain how I will market to these segments , how I see that in future , and I did that and he was happy . Then he said – you look like you have a ML or analyst background , cause all this work is done by analysts in flipkart , why management , and I took back on this question by saying that if this project ML part was developed by someone else and I just made the strategies , would he have let me sit in the interview , to which he exclaimed baat to sahi hai tumhari , sahi baat hai .
Imo interview went very well , pretty much like a quick 40 minute chat , cause I had good business acumen . I suggest my juniors who are preparing for business roles to read finshots and grow articles daily .
After that as soon as I walked out I was informed that my infoedge ds offer letter has been finalized and I cant sit for slot 2 , so I couldn’t move ahead with flipkart , to which I happily agreed and went back to hostel . (Lets just say for the record I got escorted out XD)
You will often find me saying ki infoedge nahi hua hota to definitely Flipkart hota
So takebacks – prepare project well , and have a good business acumen for this role , tech wont be asked .
One line answer - Andrews NG course , Kaggle course, Krish Nair youtube channel , statquest, and lots of practice by solving standard datasets :) Also cassendra and mosaic from Udyam (ECE dept fest IIT BHU) (Tho beware that they just surface level touch the topics and they are not exhaustive at all , do not think of udyam as the whole syllabus !!
At this point I might have given link to this repo to about 50 ppl , so might as well clear up some clutter , some of the things Ive explained below are kinda philosophyish about how to crack interviews, its what worked for me
First some general interview tips - (kya padhna hai sab bata denge thoda philosophy time :)
"You don't sell just qualities of product, you sell emotions and stories too" - Marketing is how your resume looks, how you present your github and linkedin, how you make your college life story a convincing career story that makes the recruiter hire you. Sales is how you talk to the interviewer, how you solve problems, how you behave when you cant solve a problem, and how passionate you appear about being in the field you are interviewing for and in the work of the company you are sitting for. "Its all a marketing and sales game"
When you sit in DS interview, be passionate about AI ML , when you sit in SDE interview , be passionate about serving lives of millions by making a button red or blue (Yes I love to diss SDE pplz). More specifically when you sit for navi, be passionate about fintech space , when you sit for media net be passionate about marketing, when you sit for infoedge be passionate about using data science to build models that serve huge indian audience in career, real estate, matrimony. Passionate face and personality leave a positive impact on interviwer whether tech interview ho ya hr . Ive seen SDE people who think HR Interviews are useless, but its not like that. Imagine I have 3 "perfect" candidates to choose from, I would take the one who appreciates my company and has some burning desire to work in that specific domain in which company works. Thats also why company research is also important. Since you will walk in one interview room saying that "you love data science and its powers" and enter another interview room saying "management has been your passion ever since" XD , sometimes you will end up wondering what you really like too after hearing your own self talking , thats a personality crisis , dw it cures in the end :)
Q How to learn data science -
A - My journey - andrews ng , then kaggle , then udyam (cassandra + mosaic) , then statquest , then krish nair , plus robo club and cops ig camps and inter iit for research papers, and sometimes random googling , and once iitg camp also
What I recommend -
1st 2nd yearite - entering data science - andrews ng for learning basics , kaggle mini learn courses for learning the coding part (This is very good sourvce I think) , for iit bhu people - udyam ka cassandra and mosaic will give you community in your own class , will give you a project and also a achievement to put in resume . Also need to know ML coding , need to build basic projects yourself on ipynb. start exploring Computer Vision cv , Natural Language Processing nlp , Reinforcement Learning rl, federated learning fl, generative ai genai etc maza aayega tumhe bahut. Dont stress about learning each and every command of numpy, pandas, matplotlib, sklearn, seaborn, tensorflow etc but get a surface level idea using any youtube video.
3rd 4th - knows ml , needs to do for placements - statquest for intution of the algorithm , then go deep into maths using krish nair or campusx videos , then MAS ml quizzes if you are in MAS . Also do probablity, statistics (from RICE University courses + STAT110) , linear algebra (god knows kaha se kare ye, maybe 3b1b youtube channel has good videos, or khan academy), sql (answered below) , excel (random youtube videos)
Q SQL -
A - first play SQL island game , 2 3 hours ka game hai , good for intro , then kaggle course sql ka (can skip it, but I did it so said), can see MAS notes for SQL theory, then interviewbit (uspe bahut easy questions hai even hard ones are easy), then leetcode 50 sql vaala , then MAS ka hackerrank vaala list (decent tough that uska medium bhi i would say) then datalemur (ye sabse hardest hai but its like archive of company questions) . SQL 2-3 din me nahi hota, there is lot to learn and lot of coding questions to practice in that too if you want sure short success .
Q How to select the 50 companies ?
A - Dont be the person who is asking for limit extension at the end of placement season !! Choose the 50 using this -
- Have some preference order for the domains mentioned above. This depends on your intrests and skills.
- see company size on linkedin. big companies may not allow you to work on large projects immidiately but would give you stable jobs and stable pay, startups have higher pays usually but instable , sometimes (in my senior batch this happened) companies close down too leaving you jobless even before joining. Its your decision which side you like more - stable culture or startup culture .
- talk to seniors and tpr , and get to know the past number of people hired and the difficulty of papers . For example - I personally never did cp so I avoided companies like zomato , phonepe, meesho, medianet sde where these companies are known to ask hardest coding questions .
- Pay (I mean obviously)
- One thing - Dont fill all the good companies ? Why ? you should divide the companies as per slots in which their interview happens and then pick few companies out of each slot. Personally - In slot 1-1 I had infoedge qualcomm sprinklr , and I could only attempt infoedge interview. So all the other ones I had in 1-1 were waste. You do not want the situation where you have all the company shortlists of day 1 (say you have 10-15 shortlists) and do not have anything on other days , cause its the same thing as having 3-4 companies only . Divide and conquer
Interview are of 4 types -
- directed - where your "tell me about yourself" and resume determine the next question. The next question is always made of some keyword in ur past answer or ur resume . You can easily direct this interview to your strong areas.
- Just woke up - where interviewer got to know 1 day ago that they have to take interview, they bing - "top 50 interview questions on ml/stats/sql" and they ask it. Simple defence - you google the same thing a week before.
- Rapid fire - where the interviewer has made up their mind that they want to judge you on certain topics, and come prepared with a set of questions to ask .
- torture - Where they will grill you with very hard questions purposely (even they know you cant solve all of it) and see whether you break down and fuck up or you answer the rest of moderate questions or continue with calm attitude . In extreme case they might tell you that you are wrong to test your attitude. Usually a manipulation trick used by senior interviewers.
Q Projects ?
A - Do not think that you need very high level state of the art machine learning models, making a chatgpt as a project is not good if you cant defend each and every single thing about the project. Take a moderate level project. You can see the projects I have put on my github , and my resumes are uploaded in this repo too. Inserting a github link with a good repo shows authenticity and your confidence in your project, you can also read my flipkart amba interview experience to see how I used my github in the middle of interview to explain which showed my genuineness . 2 Projects is a good number, avoid copying projects , specially when you havent done anything related to data science. Simply put - agar 2 projects bhi nahi banaye puree college life me, agar tech acha nahi lagta to tech college me seat kyu waste kar rahe ho, college life sirf g**** fukne ke liye thodi hai , its a place to grow your career .
Q Resume -
A - Uploaded to this repo - 2 projects , 1 por , some extracurricular to make you look like a human. Projects and internship ki story acche se pata honi chahiye.
Update - Since more juniors asked about resume here are my tips -
- White space kam rakho right side me to better rahega , each space is a chance to pitch urself
- I personally prefer - 2 projects 1 internship 1 POR rest achievements
- JEE rank is not an achievement, go win some branch fests
- As I said above story ready honi chahiye. Jo models project me use hue hai uspe bhi question pucha ja sakta hai
- Project link daalna is very cool
- Our TPC portal allows us to create bullets, bold things, add links etc use this feature
- I personally dont prefer to show cpi cause I have other achievements to show and cpi se konsi hi company judge karegi , infact counter question aayega ki u are good in core why data science ?
- internship kisi aur domain ki bhi ho to bhi chalega as long as company name is good and u can represent it as a part of your story.
Q How to win GDs ?
A - MAS had 10 GDs, out of which I played 9 and won 7 , so I think I can answer this.
- The first 3 people to start are seen by interviewer the most and given brownie points.
- The concluder , the person who speaks last is given brownie points . The starter, the person who speaks first, should say his name, then introduce the topic, and then give a general brief about the topic and then start speaking their views on the topic.
- If you speak, try to speak for more than 20 seconds.
- If you don't even understand the topic, you can act as "moderator" , its a person who says "we are diverting from the topic lets discuss this", "now that we have discussed benefits lets discuss drawbacks" etc .
- If you have spoken atleast twice, pointing to the person who hasnt spoken gives brownie points for kindness : "Guys I think he/she has not gotten a chance to speak, lets give them a chance , what are your views on it " etc .
- Cutting someone speaking is considered rude, so cut people , but in the least aggressive way (Like my mentor said once "If you have to be a polite person and let someone speak, might as well let them take the job too. Why sit in the placements then. ")
- My personal favourite tip - destructor - speak within the first 3 people , and put out all the points that are possible for a certain topic (this speed of impromptu thinking comes with practice) , this leaves ur opponents defenseless as anything they speak is considered as a repeat point and you win the GD straight.
- Also , when everyone starts to repeat the same point (eg , saying "yeah I agree" ), its called as a boombox or echobox, and breaking such a condition by changing the direction of discussion can grab a lot of points.
- Avoid speaking in hindi.
- Avoid using negative points like - "I think you are wrong" etc .
- Read finshots and groww articles for GK if you are keen on business roles
- This should be more than enough tips to play destructively and win fast. Smooth execution comes with great practice tho.
Q Should I join MAS ?
A - See , jiska ho jaata hai usse sab sahi hi lagta hai and ha hi kahega , and jiska nahi hota accaha vo na hi kahega . So you should see all their offerings and decide for yourselves. Here are my reviews on their offerings - They offered me 10 mock GDs which helped my impromptu skills, 5 mock interviews to polish any question answering style mistake if any. They gave me lots of practice question modules. And most importantly they knew exactly what is the "syllabus", what is asked and what isn't. And they were always availaible for discussions, as they were our seniors only. And they also called alums which they knew which were selected in the companies so that they can guide us about the tests . They also help you make resumes , and hr questions ke answers (a bit like teacher ke pass jaake answers check kara sakte ho and they will suggest modifications xd )
Q Do they ask DSA ?
A - No , infoedge does not ask DSA . Navi, Sprinklr, Hilabs, Publicis Sapient ask DSA , Infoedge, FPL Technologies don't ask DSA, and Im not sure about Microsoft and Walmart since they were on hiring freeze in my year. You are expected to know about 80-90% the DSA sde people know (Even have to do standard graphs and dp questions also). It takes about 6 months to master it so if u have time start accordingly, else start ASAP. Coding questions are asked in the first round, whether it is asked in interview or not should be asked to someone who gave those company interviews. You do not want to end up eliminated in first round just bcoz some SDE person did the coding questions and got ahead. So do prepare . Also a bit extra preparation won't hurt as it will allow sde to become ur backup.
Q Does cpi matter ?
A - Yes. How much is needed ? That I cant say, but atleast above 7-8 . When companies have too many perfect scores, they sort the list of students according to cpi and then select the top students. So ur cpi needs to be better than ur batch .
Q Internship (IIT BHU ) in data science
A - I don't think zyada companies aati hai for data science so prepare sde backup, mera 2nd week day 1 me nvidia ho gaya tha so Im not aware of Data Science scene for internships in iit bhu . Microsoft DS aayegi (tbh thoda random hire karti hai atleast mere year me , no offence) and mastercard AI aayegi. Ive heard from juniors ki Infoedge and Stanc has also started coming so u can ask ur tprs for more details (infoedge join karte ho to do ping me on linkedin using message while connecting feature :)
Q Analyst as backup role
You would need to practice SQL (sources mentioned above), Excel (thoda bahut , just pivot table and basic formulas like vlookup, sum etc), power bi / tableau (koi bhi ek event me dashboard bana lo ho jayega practice), case studies and guesstimates (iit bhu biz club posts casebooks, and every IIM posts casebooks that can help you, basic frameworks yaad karlo for standard problem types like profitability, market entry, rca etc)
Q Kitna time lagega ?
Recently I've started receiving dms where ppl are new to DS and 3-4 month before placement tests they suddenly realize DS kar lena chahiye (not targetted , 3,4 aise DM aa chuke hai) Please understand ki DS ki hiring me most companies interview for decent 4-5 hrs so unlike SDE u can't fake it or just follow 1 single thing (like sde vaale logo ka dsa karke ho jaata hai mostly) and these companies hire so less in number that they actually care about what kind of people they hire (Unlike some companies who hire half via diversity and casually fire 12k ppl, iykyk XD), so just let DS be for the people who have actually studied ML since long . Don't waste number of companies u can fill , last 3 month me u are not suddenly gonna emerge as campus ka data science expert . Take decent 6 months if you are weak in ML , or 3 months if you are already good at ML and just need to get interview ready , or 1 year if you have never even made a linear regression or CNN model .
Python - koi bhi club se karlo
SQL - mentioned above
Probs , Stats , Linear Algebra - college ka MA201 course , ya stanford harvard (not sure konsa) ka STAT110, plus RICE university course
ML - Andrews Ng, Krish Nair, Statquest , and some of my seniors did CSE229 vaala andrews ng ka original , and my juniors also did CampusX which I scrolled once and looks good to me (CampusX wasn't that much popular when I was in 2nd year, maybe its a new uprising channel)
Data - Cassandra Udyam (ECE dept fest) ke notes were good but not sufficient - https://github.com/monako2001/Cassandra_Workspace (This repo was made by my senior Aman Mishra who went into Navi)
Coding - see random youtube videos for np, pd, plt, sns (hope u know what it means), then kaggle learn modules
DSA - go to LC , throw a stone, it will hit a sde , ask them . imo , gfg ke articles and leetcode and maybe interviewbit
ALL THE BEST AND STAY IN TOUCH :)