My Interview experience at Google 2025 for MLE-III (L-5)

I was contacted by a Google recruiter via LinkedIn for an MLE-III position in Hyderabad, India. After receiving preparation materials and guides, my first round was scheduled for July 23rd, 2025.

1. Online Coding Interview (Round 1)

This 45-minute Google Meet interview presented me with a problem involving a multiline log of friend connections with timestamps. The goal was to find the earliest timestamp when everyone became friends with each other. The log could be stored in memory. I immediately recognized it as a graph problem. My first approach involved preprocessing the log to identify unique individuals, assigning them unique indices, and then constructing a graph. I planned to treat every individual as a separate connected component and decrease a global component count as I traversed each log entry and added edges between nodes. The timestamp at which the component count reached 1 would be the answer. My second, more refined approach, utilized the Union-Find algorithm. I initialized a parent array where each person was their own parent. While traversing the logs and performing union operations, I kept track of the component count. The moment it became 1, I returned that timestamp.

2. Online Coding Interview (Round 2)

In this second 45-minute Google Meet coding interview, I was given a list of lists of words, where each nested list represented a sentence. I was required to implement a method to predict bigrams. For each query with a single word, I had to return the next word that most frequently follows it in the data. For example, for input [['I', 'am', 'king'], ['I', 'am', 'travelling'], ['I', 'like', 'mangoes']] and the query 'I', the method should return 'am', since 'am' appears twice after 'I'. Unfortunately, I struggled to come up with the solution in time. I overthought the problem, assuming it would be more complex, and overlooked the simple approach of using nested maps: the outer map would store the first word as the key, and the inner map would keep a count of each word occurring after it. For each prediction call, the solution should look up the next word with the highest count.

Result

I was unfortunately rejected after the second round. The recruiter called and informed me that I did not perform well in the second round, and they would not be moving forward with my application.