[Uber - L5A] SSE interview experience

I appeared for the Uber hiring drive around mid-May. The interview process consisted of five rounds, covering data structures & algorithms, machine coding, system design, and behavioral aspects.

Round 1: Screening (DSA)
This round felt like a proper DSA assessment where a working solution was expected. The question was: Given an n*m 2-D matrix with arrows in different directions, find the minimum cost to travel from the top-left grid to the bottom-right grid. Traveling in the direction of a grid arrow costs nothing, but moving in any other direction costs 1 quantity.
I initially suggested a DFS approach, mentioning it was brute force, and we discussed its time complexity. I then proposed a graph-based approach, explaining that I would construct a weighted (0,1 edge) graph and apply Dijkstra's algorithm. I managed to code this within 60 minutes.

Round 2: DSA
In this round, I was asked to find the next greater palindrome number for a given number. I had encountered a similar problem before, which allowed me to write a working solution within the hour using a 'Mirror and Carry Propagation' approach.

Round 3: Machine Coding
This round asked me to implement an expiry counter. I needed to create a class with methods like put_element(k), get_element_count(k), and get_total_elements(), initialized with an expiry time 't' seconds. I initially suggested a priority queue and hashmap-based approach, but the interviewer wasn't convinced. I then proposed a binary search based approach, maintaining a mapping of elements to a list of timestamps (e.g., 'k': [t1, t2, ...]). For reads, valid timestamps would be found in O(log n). I coded this in Python, and it worked, but the interviewer questioned why I wasn't discarding stale timestamps more efficiently, which I believe led to an initial 'NO-HIRE' feedback for this round.

Round 4: System Design
This round, taken by a Staff Software Engineer who was also the bar-raiser, focused on designing a heatmap of Uber drivers in a city for an internal analytics dashboard. I needed to cover two scenarios: a near real-time heatmap for the past 20 minutes and a dashboard showing hourly buckets after 24 hours. I had some prior exposure to this problem. My approach for the real-time use case involved consuming location events from Kafka, using a geo-hashing library, pushing to a Kafka stream partitioned by geohashId, and using an Apache Flink cluster to count over a 1-minute tumbling window, persisting to Redis. For the 24-hour use case, another Flink processor would flush hourly aggregates to PostgreSQL. The interviewer had minor questions on scalability and fault-tolerance, but overall, the feedback was 'good'.

Round 5: Behavioral/Leadership (TLM)
The final round was with a Tech Lead Manager. After discussing my current and previous roles, I was asked to detail a complex or impactful project I had worked on, focusing on both technical and non-technical challenges. This was a 45-minute monologue with minimal interruptions. We also discussed the type of work to expect at Uber and my career aspirations.

A week later, I received a call from the recruiter informing me that despite the initial 'NO-HIRE' in one round, the HM and BR had overturned it in the final debrief, and I would be receiving an offer!