I applied through an employee referral. The process took 3 weeks. I interviewed at Google (San Francisco, CA) in Dec 2020
Interview
5 interviews, 1 call with a recruiter, followed by technical interview then three on-site (now hangouts). The call went well. The interviewer was about 10 minutes late and so he didn't even do introductions and stuff. He jumped straight to questions. Fortunately, the 4 questions were not very hard so I finish just in time.
Interview questions [1]
Question 1
Basic to intermediate SQL questions on windows functions and nested queries
30 minute phone screen with HR, followed by an interview with the hiring manager. HR would not even provide a salary range for the role, which was very weird. The HR rep was not familiar with the role and seemed to be reading from the JD when I asked questions about it.
I applied through an employee referral. The process took 2 months. I interviewed at Google (Seattle, WA) in Aug 2021
Interview
Recruiter screen > tech screen > 5 tech sessions at remote "onsite"
Tech screen: all statistics written in easy python
On-site: python for SQL-style queries, one session focused on stats/probability, majority of sessions had some probability in it, some question were extremely open ended, hierarchical statistical models, optimization and creating penalty functions, bootstrapping, small sample statistics
Interview questions [1]
Question 1
you are given a discrete probability distribution of children, what is the probability a random women you meet on the street has a sister?
Two variables x1 and x2. They are correlated but aren't the same. X3 = X1-X2 and X4 = X1+X2. What are the coefficients for x1 and x2 if you train logit for x3 and x4
1000 ad videos, 1000 human raters
Assess the quality of videos, 100 randomly selected videos to each rater, Rate video between 1 (bad) and 10 (good) quality. How would you rate these? What are the pros and cons of your strategy?
clustered statistical modeling question about how you would set data up for this model and what model you would use.
