Uber L4 Software Engineer Interview Experience Accepted

Candidate Profile

Experience: 5.5 Years

Company: Uber

**

Round 1 Screening** *

Problem: Find the Closest Palindrome (LeetCode Hard). *

Verdict: Yes **

Round 2 Data Structures & Alg

Candidate Profile

Experience: 5.5 Years

Company: Uber

**

Round 1 Screening** *

Problem: Find the Closest Palindrome (LeetCode Hard). *

Verdict: Yes

**

Round 2 Data Structures & Algorithms** *

Problem: Given a grid containing police stations, one thief, and one bank, determine if a path exists for the thief to reach the bank without being caught. Each police station patrols within a fixed Manhattan distance ($k$). *

Solution: Implemented Multi-source BFS to mark danger zones defined by police patrols, followed by DFS to validate the thief's path. *

Follow-up: Solve for a scenario where the patrol distance ($k$) varies for each police station. *

Follow-up Solution: Implemented Dijkstra’s algorithm. *

Verdict: Strong Hire

**

Round 3 Low-Level Design (LLD)** *

Problem: Design a data structure with Time-to-Live (TTL) functionality for (key, timestamp) pairs arriving in increasing order. Required features include: 1. Count of all active keys. 2. Count of active keys for a specific key. 3. Standard CRUD operations. *

Candidate Approach: Used a Map<String, Queue<Timestamp>>. Expired keys were removed by iterating through the map during function calls. *

Optimized Approach (Suggested by Interviewer): Use a Queue<Pair<Timestamp, Key>> for expiration management, coupled with a Map<String, Integer> for specific key counts and a global variable for the total count. This allows directly polling expired elements from the queue based on timestamp without iterating through map keys, resulting in better average-case complexity. *

Verdict: Lean Hire

**

Round 4 High-Level Design (HLD)** *

Problem: Design a recommendation system for the Amazon homepage. *

Solution: Modeled as a Top-K problem. Utilized Flink to track metrics (item views, purchase counts) and maintained sorted sets to generate real-time recommendations. *

Feedback & Challenges: * Scale estimations were imprecise. * Metric selection was limited (focused primarily on views, buys, and ratings). * Failed to account for localization (e.g., maintaining different datasets for different countries). * The architecture used separate Redis caches for different metrics, limiting extensibility. *

Verdict: Hire

**

Round 5 Hiring Manager** *

Focus: Behavioral questions regarding past professional experiences and responses to hypothetical workplace scenarios. *

Verdict: Strong Hire