Google

PART 1: ALGORITHMIC PROBLEM - DISTRIBUTED TASK SCHEDULER BASE PROBLEM You are building a build system that needs to compile 'N' different software modules (labeled 0 to N-1). You are given an array of 'dependencies', where each dependency is a pair [A, B], meaning module B MUST be compiled before module A can begin. Task: Write an algorithm to determine if it is possible to compile all the modules. If it is, return a valid order in which they can be compiled. If it is impossible (e.g., due to a circular dependency), return an empty list. FOLLOW-UP 1 Now, assume every module takes a specific amount of time to compile. You are given an array 'times', where times[i] represents the compilation time of module 'i' in seconds. Assume you have an unlimited number of servers, meaning you can compile as many independent modules in parallel as you want. Task: What is the algorithm to find the absolute minimum amount of time required to compile all 'N' modules? FOLLOW-UP 2 In reality, server resources are not infinite. You are now constrained to exactly 'K' concurrent worker servers. This means at any given time, a maximum of 'K' modules can be compiling simultaneously. Task: Discuss how you would approach this new constraint. How does limiting the workers change the complexity of the problem, and what data structures would you use to simulate this process and find the total compilation time? PART 2: AI DISCUSSION & TERMINOLOGY Practical Application & Security: When working with proprietary company code or sensitive user data, what precautions should you take before using public AI assistants to debug or format your work? General Knowledge: What is the difference between "Supervised Learning" and "Unsupervised Learning"? Can you give a brief, real-world example of each? Basic Terminologies (Briefly explain the following concepts): Fine-Tuning: The process of taking a pre-trained AI model (which already knows general language or concepts) and training it further on a smaller, specific dataset so it becomes an expert in a particular niche (like legal documents or a specific coding language). Prompt Engineering: The skill of structuring text inputs (prompts) given to an AI model in a way that maximizes the accuracy, relevance, and formatting of the output. Embeddings: A way to translate text, images, or audio into lists of numbers (vectors). This allows the AI to measure how mathematically "close" or "related" two concepts are (e.g., the embedding for "dog" will be closer to "wolf" than to "car"). Parameters: The internal variables or "weights" that a neural network adjusts during training. When you hear about an AI having "70 billion parameters," it refers to the massive number of these internal connections that hold its learned knowledge.