Google

# Practice Interview Set 82 ## Part 1: Algorithmic Problem — Intervals & Sorting ### Base Problem: Merge Intervals You are given an array of `intervals` where `intervals[i] = [start_i, end_i]`. Task:** Merge all overlapping intervals, and return an array of the non-overlapping intervals that cover all the intervals in the input. Why is sorting the intervals based on their `start` time the critical first step? What is the time complexity of this sorting step? Once the array is sorted, how do you iterate through the list in exactly one pass ($O(N)$ time) to compare the `end` time of the current merged interval with the `start` time of the next interval? - ### Follow-Up 1: Insert Interval You are given an array of non-overlapping intervals that are *already sorted* by their start times. You are also given a new single interval, `newInterval`. Task:** Insert `newInterval` into the intervals, merging if necessary, and return the updated array. Since the initial array is already sorted and non-overlapping, using the $O(N \log N)$ sorting approach from the Base Problem is inefficient. How do you construct the new array in strictly $O(N)$ time by breaking the process into three linear phases: (1) add all intervals occurring *before* the new interval, (2) merge all overlapping intervals into the new interval, and (3) add all remaining intervals occurring *after*? - ### Follow-Up 2: Meeting Rooms II You are given an array of meeting time intervals where `intervals[i] = [start_i, end_i]`. Task:** Find the minimum number of conference rooms required so that no two meetings overlap. *Approach 1 (Min-Heap):** After sorting by start times, how can you use a Min-Heap (Priority Queue) to keep track of the *end times* of currently active meetings? Why does the size of the Min-Heap at any given time represent the number of rooms needed? *Approach 2 (Chronological Ordering):** Alternatively, how can you separate all `start` times and `end` times into two independent arrays, sort them both, and use a Two-Pointer approach to track active meetings in $O(N \log N)$ time and $O(N)$ space? - ## Part 2: AI & LLM Core Concepts (Very Light / Foundational) ### Question 1: Supervised vs. Unsupervised Learning In the world of machine learning, what is the basic difference between **Supervised Learning** and **Unsupervised Learning**? In plain English, what does it mean to have "labeled data," and which approach is typically used when you want an AI to categorize photos into "Cats" vs. "Dogs"? - ### Question 2: AI Bias and Training Data Sometimes, AI models output text or generate images that reflect human stereotypes (e.g., assuming a doctor is male and a nurse is female). Why is this generally considered a "Data Problem" rather than a traditional "Coding Bug"? How do AI models inherit these biases? - ### Question 3: Time-to-First-Token (TTFT) and Streaming When building software that uses an LLM API (like a chat app), developers care deeply about a metric called **TTFT (Time-to-First-Token)**. What does this mean? Why do modern AI apps stream their answers word-by-word onto the screen rather than making the user wait for the entire paragraph to finish generating? - ### Question 4: The Role of "Hugging Face" If you read tutorials about building AI software, you will constantly see references to a company and platform called **Hugging Face**. In simple terms, what is Hugging Face, and why is it so frequently referred to as the "GitHub of AI" by software engineers?