45-minute phone interview: Program a computational geometry problem. The phone interviewer did not ask anything else at all.

onsite:

preparation: I prepared for two weeks. I had been asked to brush up on my algorithms, so I basically took the online versions of Prof. Tim Roughgarden's Algorithms class at Stanford and parts of Prof. Erik Demaine's class on Advanced Data Structures at MIT. GeeksforGeeks.org also helped me. Even if you don't get into Google, I would say that taking these classes is very good for your future technical skills.

Five people interviewed me, each for 45 minutes. None of the five interviewers asked me any algorithm-heavy questions. There were no TopCoder-style questions either. Most of the questions were related to my domain area, viz: computer vision and machine learning. Since I had just switched into these areas, I was not an expert, and my answers were often patchy. One of the interviewers asked me a question on large-scale system design.

I am just waiting for the results.

An open-ended question on large-scale system design. For this, you need to be an internet programmer, knowing about issues related to data replication, data availability, large-scale scaling, etc. I was very patchy on all of these areas.