1. Coding round and basic Maths and ML questions. 2. ML round with deep level questions, varying on interviewer. They may ask everything from LA, Probability and ML, DL. 3. Managerial round with some case study level question.

Round 1-Tech ~ 35 mins What is multi collinearity, how do we detect it. What is the actual problem with having features that are multi collinear ( Coz our goal is hitting higher accuracy)- Read about it properly . The issue is interpretability of the coefficient values as the variance of their estimated value gets inflated due to multi collinearity They grilled a lot regarding the multi-collinearity topic. First, they asked me to explain mathematically . When I did, he said I am not interested in maths , explain intuitively – So keep both aspects ready Stochastic gradient descent algorithm, it's convergence In general discussion of gradient descent What can be improvement over this approach ROC- AUC Curve , what is its significance – Read about what the area under it represents and explain with example like –if the AUC is 0.8 it means that there is 80% chance that the model will assign a higher probability score to a point belonging to +ve class , comapared to a –ve class point Some questions about neural networks ( ANN , Loss functions , overfitting issues )- They will want to hear about L1,L2 Reg. So read them properly Explain how the addition of L2 norm to the loss function reduces overfitting mathematically How do we get the Q,K,V in a transformer architecture explain from the start where we pass a sentence – You need to explain tokenization , vocabulary formation, one hot vector for every token , embedding matrix getting trained and then word embeddings What is the improvement of transformer architecture over recurrent neural networks How is the positional information is getting retained in that context Multi headed attention , how the weights and embeddings flow sequentially What is masked attention , where do we need , how do we achieve it mathematically Questions regarding the transformer models I was using in my Summer internship , its architecture, Difference between encoder based models and decoder based models Then I was asked to share my screen and write a code for sorting an array and extract the 3rd highest element ( without using arr.sort() kinda commands) Verdict- Cleared Pro tip: Read properly and develop the depth of knowledge in the topics coz in full time placements, merely yapping AI will not do the job Round 2- Technical Case Study ~ 25 mins It was a case study on how to optimally allocate resources( Like establishment of towers , air fibre lines or any other resources ) - in a region. The target is Which areas should I focus on – in order to do that , what kind of analysis you will do , what kind of problem statement you will frame , what kind of features will you include etc Its an open ended question – you can search similar questions in any platforms for interview prep . One thing they really love is to translate these kinda problems into a hypothesis testing framework and then establish using t test Verdict- Cleared Tip: Just practice the general structure of DS case studies Round 3- HR ~ 20-25 mins Hobbies , Life , What is the most difficult situation in your professional life you have been to , How did you overcome Why are you interested in Airtel How will you resolve conflict within your team How will you Communicate technical details to non-tech stakeholders Some others I don’t remember Verdict- Did not make it to final list.