It absolutely sucks to be rejected **yet again** but we just have to keep chugging along and improving. This was my first "on-site". I\'ve been rejected by many FAANG interviews over the years before getting to this round so I know I\'m getting closer to that elusive offer. It hurt not being able to type [Offer] instead :\'(\n\nOne major difference in my prep is using this website. In previous years I used to study data structs and learn the nifty algorithms like KMP / Knapsack etc. This was clearly the wrong way to approach the technical interviews. Doing the problems on Leetcode and the 30-day challenges with a contest here and there really helped me track my progress. As the weeks went on, the problems got easier and I was able to do them faster. Over time you\'ll pick up tips and tricks on handling various problems and you\'ll be able to spot them easily in interviews. \n\nPhone Interview 1: https://leetcode.com/problems/word-search-ii/\nPhone Interview 2: Variant of https://leetcode.com/problems/two-sum/\n\n# **Virtual On-site:**\n**Interview 1: ML + Algorithms**\nLots of LP questions\nTalk about research process and how I handle working with various teams\nTalk about scalable ML and how ML looks in prod\nCoding problem: something similar to\nhttps://leetcode.com/problems/edit-distance/ \n\n**Interview 2: ML**\nIndepth ML discussion\nTested ML Breadth with pros and cons of various models for various situations, datasets, objectives\nPros and cons of various loss functions\nLots of LP questions\n\n**Interview 3: Bar Raiser**\nThis interviewer was extremely nice! Clearly far more experienced than most other interviewers both at Amazon and all of the other places I\'ve had interviews. It felt like a friendly light conversation with lots of positive feedback.  \nLots of LP questions.\n\n**Session 4 - Tech Talk:**\n30 minute presentation of my PhD work followed by 30 mins of Q&A focused on ML. \nWhy use certain models and why tune certain params etc. \nJustify the decisions made during the research.\n\n**Interview 5: ML with Hiring Manager**\n30 mins LP questions\n30 mins technical ML questions. \nDive deep into a few models to test my understanding of the underlying datastructures and why certain decisions are made when constructing the underlying structs. \n\n**Self ratings**\nInterview 1 went extremely well. I knew all of the technical questions and solved the problem using the optimal time/space complexity quickly. I rate my performance on this 10/10.\n\nInterviews 2 and 5 were rough. There was plenty I didn\'t know and plenty I did know. They could tell where I was unsure and drilled into it to expose my lack of knowledge in a very specific topic. I rate myself 6/10 for each. \n\nInterview 4 (bar raiser) is hard to call. Given that it was mostly LP questions and I felt like we both enjoyed the chat, I\'d say it went well. I didn\'t have stories that fit to the questions perfectly so I had to mold some of my prep.\n\nTech talk went smoothly as I just had to read some of my papers to refresh my justifications of why I did certain things. \n\n\n**About me:**\nPhD Machine Learning (graduating later this year depending on COVID)\nMasters CS\nBSc CS\nLeetCode : Easy 80, Med 50, Hard 10\nOther algorithms websites: I completed LOTS. I like doing algos for fun.  \n\nThe recuiter said that most interviews were positive but overall you need a lot of DEEP ML knowledge for the RS/AS positions. They instead suggested I try for a ML Engineer position.

It absolutely sucks to be rejected yet again but we just have to keep chugging along and improving. This was my first "on-site". I\'ve been rejected by many FAA