Background: PhD in ML with 4+ years ML exp at non FAANG top tier company

Facebook Recruiter reached out and I had phone screen in December: Standard 2 Medium LC's from FB tagged top 100 with minor variations and I was informed next day that it went well and I scheduled my onsite in late Feb.

Virtual Onsite

• Coding 1: Two Medium LC's from FB tagged top 100 with minor variations . Interviewer seemed happy at end as I was able to code both of them and run through test cases. Seemed interested in an extension of second question to large data which cannot fit in memory. Asked to write basic functions for this extension but we were out of time so I spoke of an idea.
• Coding 2: Two Easy LC's from FB tagged top 100 with minor variations. I did the first one well adjusting for the variation but I did a slight messup in the next one which I believe cost me the interview.
• Behavioral: Standard questions and 1 LC easy - I believe this one went fine.
• System Design: Was asked a design question not from GSD course or general FB Sys Design question. I went through my design and interviewer was mostly patient to let me lead it by interrupting as and when further clarification was needed. I did put an architectural drawing in Google drawings (spoke of components in the design for availability, partitioning, consistent hashing, priority queues, etc) while talking completely throughout the interview.
• ML Design: I was asked an open ended question to design a recommendation feature for FB. I tried to explain all details to interviewer who seemed to nod his head all along. He finally mentioned that the design makes sense. He also said he was taking notes so I can continue explaining without writing/drawing the design. I had a decent connect with interviewer and I thought this round also went well. I believe I made a mistake here also by letting the interviewer take notes based on my explanation rather than me write them in Google drawings myself. (Its tough to say which option is better as each has its pros and cons)

Interview experience: Overall interview was well conducted and interviewers were polite. I felt I did well in general except for the minor messup in coding round 2 for second question. I have seen folks mention that behavioral and ML design are key for E5 in ML which did go well for me, so I was positive.

Recruiter sent a one line reject email within 3 days with no feedback whatsoever.

This is very disappointing to see a company like FB behave like this and expect the candidate to re-apply within one year when you are not giving any feedback for rejection in the first place. I have had rejections before also where a recruiter would atleast call and speak for 5 mins to go through the feedback. I have reached out to the recruiter for feedback but I am not sure he will respond. I have seen similar behavior by recruiter for another FB E5 post here, but did not expect this would happen to me.

Some feedback for folks interviewing for E5 ML

• Coding: Even a small bug means in one question out of 2 or 3 questions implies you have lost that complete round. You need to finish 2 medium LC's (with extensions if possible) and run through test cases completely by end of interview to stand a chance.
• I did not find the ML folks to be technical enough at FB, so folks with significant knowledge in ML should definitely answer the design questions in an extremely dummified way.

Google (L4-L5 ML SWE)

Advanced directly to onsite possibly due to good performance in previous onsite 3 years back.

• Coding 1: 2 LC Medium problems but not asked as a usual LC problems. Google interviewers might present it as an ambiguous question. I was able to do well here and interviewer was really nice. He gave a confirmation of the same.
• Coding 2: Very open ended coding question which was ambiguous and needed communication to understand data structures to use. Again nice interviewer and I think I did a decent job in here.
• Behavioral: Standard behavioral grind. Tried to answer naturally as I had prepared
• ML 1 : Very unusual ML question which needed to write code for a Bayesian classifier (again this was hidden in the ambiguity of the question). I resolved the ambiguity and came all the way to a Bayesian approach but could not write the code for the classifier due to lack of time.
• ML 2: Lots of discussion on CNN's and how to apply them for a real-world problem.
• ML Design: Design a feature for Google search. I did put all my thoughts in the google document while talking through but interviewer seemed a bit puzzled by my approach.

Overall interview experience: Google interview was a bit less organized than the FB interview. However, 4 out of 6 interviewers I met this time at Google virtually were really nice to talk to. I had mostly grumpy interviewers during my first onsite 3 years back so this was a welcome change. Recruiter also seems to be a very nice person. I have to see how she handles communicating the result etc.

Some words of advice

Google prep is very different than FB prep. I believe FB prep has a template and any deviations from that template implies direct rejection. So it is very important to know the FB template and then prepare, if not your hard work will be wasted. Google questions on the other hand are very ambiguous and open ended. Google definitely follows no template. Eventually hardwork alone is not enough for FAANG and you do need luck. I suppose I will know my Google outcome in the next 7-10 days given the general slow way their process runs.

My prep

I finished 350 questions on LC over 9 months. (overall LC count - 633 - 100 hard + 533 (medium + simple)). I did very exhaustive ML prep and finished the GSD course for System design for FB. Overall, I prepared very intensely for 3 months maxing out most of my day including squeezing time at work to prepare. I am disappointed with the FB outcome given my prep and recruiter response, but there are much more worse things happening in the world now. Wish everyone chasing FAANG and other top companies all the best!