Goldman Sachs | Data Analyst | Interview Experience | Dec 2020 [Offer]

Round 1: Technical Discussion - Projects

The first round kicked off with a standard 'Tell me about yourself' and discussions around my resume, specifically addressing my lack of internships. I confidently highlighted my 'mini-torrent' and 'wikipedia search engine' projects. We dove deep into 'mini-torrent', where I explained its step-by-step construction, architectural details like handling slow trackers (active/passive mode with two trackers), and hash calculation using SHA1 for both chunks and complete files.

Next, we moved to my 'wikipedia search engine'. The interviewer presented a scenario: how to manage repeated search requests served by multiple random servers. I proposed a caching approach and was asked to detail its implementation with appropriate data structures and optimizations, specifically explaining LRU with memory considerations. I felt confident in my explanations throughout this round.

Round 2: Competitive Programming - Data Structures & Algorithms

This round was purely focused on competitive programming. After a brief 'tell me about yourself', I was given my first problem:

1. Find Duplicate in Array (N+1 size, 1 to N values): Given an array of n+1 size, with values from 1 to n (inclusive), where some values are repeated, find any one repeated value. I discussed various approaches, but the interviewer was pushing for the optimal solution using the slow and fast pointer technique, which I initially struggled to articulate.
2. Minimum Rating in Streaming Data (Top N/3): The second scenario involved a website with a rating system where ratings arrive in a streaming fashion and can be very large. The goal was to display only the top N/3 ratings and identify the minimum rating among those displayed at any point in time. My initial thoughts drifted to a max heap, then a vector, but after a hint from the interviewer to consider heap data structures, I correctly pivoted to a min-heap and proposed an efficient solution.

For every competitive programming question, I was required to write code.

Round 3: System Design & Conceptual

The final round started with a light-hearted question about my interview experience. I again brought up my 'mini-torrent' project, but seeing the interviewer's lack of interest, I switched to my 'wikipedia search engine', which immediately piqued his interest. During my explanation, he interrupted with two scenario-based questions related to its architecture:

1. Handling Natural Language Phrases in Search Engine: The interviewer asked how my search engine would handle natural language processing data. After some clarification, as he himself struggled to explain NLP, I understood he wanted to know how to search for corresponding or related documents when a user provides a phrase as input. I successfully explained my solution.
2. Distributed Inverted Indexing with K-way Merge: Given a very large dataset across multiple servers, with the ability to perform K-way merge, I was asked to propose a solution for building an inverted index. I suggested implementing K-way merge sort on individual data segments on each server, followed by merging all sorted data from all servers to build the inverted index, which is a standard approach.
3. Autocomplete Suggestion (Next Word Prediction): The last question was about implementing an autocomplete feature: if a user types 'Goldman', how would my system suggest 'Sachs' as the next word? I provided a solution, acknowledging it might not be the most optimal but served as a valid starting point.

I concluded the round by declining the offer to ask any further questions, as my doubts were already cleared in previous rounds.