Machine Learning Engineer System Design Interview Questions

Precious H. - "Firstly, In designing a denoising system for sounds, I would start by clarifying the type of noise either stationary or non-stationary and application constraints which are the latency, scalability, accuracy and deployability. For real-time systems like Google meet, I will prefer a hybrid DSP + ML model like RNNoise. For batch processing like YouTube audio enhancement, a deep learn-based system such as Demucs or SEGAN would work well. Then I will need to evaluate how well the system improves aud"See full answer

Adam L. - "Discussed: Requirements of the system: latency language modality (assume keyboard typing) availability of data (assume cold start) success metric (accuracy of next word predicted?, or minimize false positives? -> accuracy to start) Data collection and processing: design ethical user experiments to collect typed out data design a simple tokenization strategy (word level encoding, character level encoding, byte-pair encodings, and discuss tradeoffs) collect data, and split"See full answer

Ajay P. - "Designing an evaluation framework for ads ranking is crucial for optimizing the effectiveness and relevance of ads displayed to users. Here's a comprehensive framework that you can use: Define Objectives and Key Performance Indicators (KPIs):** \\Click-Through Rate (CTR):\\ The ratio of clicks to impressions, indicating the effectiveness of an ad in attracting user attention. \\Conversion Rate:\\ The ratio of conversions (e.g., sign-ups, purchases) to clicks, measuring how well"See full answer

Michael F. - "Prompt: We work for an online shopping website. Our team wants to consider offering discounts (e.g. 10% off your next purchase) to customers to incentivize them to make purchases. How would you design a system that decides how to offer these incentives? Answer Goals: Increase customer engagement while controlling costs. Specifically, we want the increase in revenue per customer per week of customers that receive the discount to be greater than the cost of the discount. Metrics: Revenue per cu"See full answer

Jacky Y. - "Problem scope: Can this system detect Bot in real-time online or offline? Both. Online traffic: 1M DAU. Latency: 2s. Offline frequency: daily Offline data: 2B activity logs. Data: How do we know a Bot player (Label)? Human label. Imbalance data: reweight, resample. Develop a Bot simulator to generate more data offline for training. Given lower weight to simulator data than human label. Features: Signals from different models online. Log all the features for offline. Propose new features: dail"See full answer

Kartikeya N. - "C : Okay. So I would want to start with knowing what is the product for which we have to build a recommendation system. I : This is a photo sharing product. C : Okay. So is this something on the lines of Instagram? I : Yes C : Okay. And are we a new product co or we have some current product built already? I : You can assume yourself. C : Okay. Is there any demography or country we are targeting? I : No, this is a global product C : Okay. So, the biggest goal of any product recommendation system"See full answer

Clayton P. - "The interviewer hinted that a two-tower recommender system might be a suitable approach, using user history to embed users and pages separately and train on view or interaction data. Instead, I proposed a different approach that I felt was more aligned with how knowledge is structured in Confluence: I designed a system using a graph database to model the relationships between Confluence pages. Each page is a node, and edges represent content-based references. For example, when one article"See full answer

Sai vuppalapati M. - "At a high level, the core challenge here revolves around building an effective recommendation algorithm for news. News is an inherently diverse category, spanning various topics and catering to a wide array of user types and personas, such as adults, business professionals, general readers, or specific cohorts with unique interests. Consequently, developing a single, one-size-fits-all recommendation algorithm is not feasible. To enhance the personalization of the news recommendation algorithm,"See full answer

Scott S. - " Functional Requirements Content Ingestion\: Ingest news articles from various sources (websites, social media, etc.). Handle different types of content (text, images, videos). Content Analysis\: Extract and preprocess text from articles. Analyze the content for potential indicators of fake news. Model Training and Prediction\: Use machine learning models to classify content as fake or real. Continuously improve models with new data and f"See full answer

Anonymous Hare - "Sharing the approach for functional requirements we tool to solve this question. Functional Requirements This is only for the Registered users What is a "For You" page ? Home page where you get suggestions based on people you follow. Interactions like/share/comments (done by user) Interests (shared by the user during registration or onboarding) sports choices/ region choices/ Video sharing platform. So how many videos should we s"See full answer

Deepak K. - "FN Given text need to figure out is it following guidelines. Should notify the user in case of not following guidelines. Reason for failure should have misleading/spam/adult filters. NFN High availability High Scalability Low latency of processing Estimations 1M requests/min text - 10kb => 9.5GB/min => 14TB/day API fetchmoderationscore(text) score will be between 0 to 1 more than 0.8 => not following guidelines fetchmoderationscore(text, filter)"See full answer