This is a record of this year's walk with our trainees. Uber Backend SDE2 project. From the HM to the interview, to the four rounds of VO at the end of August, the overall difficulty is really in the intermediate to advanced level, with high frequency of coding + depth of design.
I helped him calibrate his thinking during the whole process, giving structured hints for the coding part and reminding him of the direction for the system design part, so that he could keep a clear rhythm and stable logic. The final performance of the interviewer's comments were very good, and the HM round was surprisingly smooth.
Timeline
- Early July HM interview → phone interview (scaled down LRU)
- Four rounds of VO at the end of August:
- LC 815
- HM deep chat
- LC 269
- System Design: Design a system similar to ChatGPT (big factory level)
Telephone face (reduced version LRU)
This round is very typical of Uber HF medium. interviewer valued:
- Does the data structure combination get to the point at once
- Is the complexity of the operation clear
- Realize whether clean or not, without detours
In the preparation stage, I help participants to run through the "Uber interviewer's preferred version" - that is to say hashmap + double linked list of the minimum necessary dictation model, common follow up points (concurrency, eviction policy, understanding herd), and which points cannot be made too complex.
On the day of the phone interview we kept up the pace with voice reminders - for example, when he was getting ready to implement, I whispered next to him, "Write the node class first, keeping the method order get → put."
It ended up being a very smooth write and took no more than 15 minutes to take.
VO Round 1: Coding (LC 815 - Multi-Source BFS)
This is the type of question Uber really likes in the BFS/graph direction.
We reminded participants before we got into coding that Uber is more concerned with your problem reasoning than with straight code.
So the trainee structured the points to make them clear:
How the graph is built (bus → stops or stops → bus)
BFS Why it's right
Whether access to de-duplication is required
Explanation of Time Complexity O(N + M)
When he got to the third step, I added a line via voice prompt for him to ask the interviewer, "Do you have a preference for building graphs at bus-level or stop-level here?"
This phrase is crucial to make the whole exchange more natural and to make the interviewer feel that you are the kind of engineer who "works with the interviewer to solve problems".
The code parts are cleanly written and the test cases are going by solid.
VO Round 2: HM Deep Chat
This round focuses on mindset, teamwork, and a solid understanding of the business.
I went ahead and gave him two mocks of the categories that Uber HMs often ask about:
- How to lead complex problem deconstruction
- How to drive cross-team collaboration
- How to do decision-making trade-off
- How to Explain Technology to Non-Technical People
It's all been trained, especially the "storytelling to the point of specificity and landability" point.
The examples he ended up telling in front of HM were very engineered, from impact → bottleneck → solution → metric, a set that HM was clearly satisfied with.
VO Round 3: Coding (LC 269 - Hard topology)
This is not a difficult question and very examinable:
- How to build a directed graph from string relations
- How to determine the invalid case (prefix edge case)
- Oral stabilization of topological sort
The trainee was a little nervous about entering the room, and I gently reminded him via voice: "Don't think about the code yet, draw the example first, then say how the diagram will be built."
That statement was crucial and steadied him in his rhythm.
Throughout the round, his thoughts were very much like the "mature backend engineer" version:
- Build map → in-degree array → BFS
- Checking for prefix invalid
- Need for processing in case of multiple answers
The interviewer basically didn't question his logic.
VO Round 4: System Design (design a system similar to ChatGPT)
This is the most critical round and the one that Uber values the most.
We press directly in the preparation phase Uber/Meta level system design The standard gave him four intensive mocks, including:
- Requirements validation (form of user input, response latency targets, concurrency scale)
- Request path derivation (API → Gateway → Orchestration → Model Serving → Cache → Feature Store)
- How to choose between high availability and high consistency
- Cost awareness (cost of LLM inference)
- How to disassemble P95/P99 latency?
- Scalability, how do you expand it?
On the day of the formal interview, he was very solid in his presentation of the whole process:
The main line he was telling the interviewer was probably:
- API and request structure defined first, sync vs async clarified
- Inference server how to slice / split model weights
- How Token streaming is implemented
- Add queuing & batching before model server to improve throughput
- Cache (embedding / previous conversation context) how to do it
- Bottleneck: GPU saturation, context length, load peaks
- Solutions: dynamic batching, multi-cluster, failover
He completely dictated the pace for the entire 45 minutes and was able to push through every follow-up question the interviewer asked.
I only reminded him gently when he confirmed the requirement: "mention the P99 latency target" and "ask about multi-region".
The interviewer's attitude became noticeably more positive after these two questions were asked.
Programhelp Interview Support - Hands-on Passing Support
The case of Uber SDE2 is actually the typical way we bring students in our daily life: instead of simply giving answers, we benchmark the real listening habits of interviewers in big factories, and adjust the expression, rhythm, and decision-making process of the students to the level of engineers.VO On-site, we pay more attention to the rhythm and point of landing to ensure that the answers are not scattered or stuck, and the overall presentation of the way of thinking of mature engineers. The entire process is seamless, secure and safe. If you are also preparing for similar big factory interviews, you can tell me the position and progress, I will give the most suitable program according to your situation.
