Just finished last month Lyft The entire process of Software Engineer Intern / New Grad, and finally got the Offer. The whole process took about 5 weeks. The overall feeling is: The Lyft SDE interview is more pragmatic and not as algorithm-focused as some companies, but it requires high engineering capabilities and systems thinking.
Below I will fully share with you the two parts of my experience of OA and VO.
Lyft OA
Platform:HackerRank
Time:75 minutes
Question volume: 2 Coding questions
The first question is Easy-Medium difficulty, which tests string processing and rule simulation. It takes about 15 minutes to write.
The second question is Medium, which is a graph + greedy question, similar to the "matching problem of drivers and passengers", where distance and waiting time need to be considered. I used BFS + priority queue to do it, and it took about 35 minutes to get through all the cases.
Tips: Lyft’s OA time is quite friendly, but the modeling of the second question is more important. If the idea is wrong and it is easy to get stuck, it is recommended to manually simulate the sample several times before writing the code.
VO - 4 rounds in total, 1 hour each
Round 1: HM + Behavioral Round
HM first briefly introduced the business direction that the group is working on (mainly related to Lyft’s destination recommendation). After that, I spent most of the time digging into my resume, interspersed with many classic BQ questions, such as:
- How did you handle conflicts with your colleagues?
- What is the biggest difficulty you have encountered in past projects? How was it solved?
This round mainly depends on personality and cultural compatibility, there are no particularly difficult questions.
Round 2: Coding round
(Specifically, it is an array + hash question of Medium difficulty). My interviewer was a Chinese interviewer. He was super nice and the communication was smooth throughout the process. He will first let me explain my ideas clearly, and then start writing code after confirming that there is no problem. After writing, he will review the time complexity and boundary conditions together.
Round 3: ML System Infra Round
The interviewer was a Senior Staff MLE, and the question was very close to Lyft’s real business: “How do you recommend destinations to users when using the Lyft App?”
He was very friendly throughout and kept guiding me in the right direction. We talked about everything from data collection, feature engineering, model selection, offline training, online serving, to AB testing and monitoring.
Round 4: ML System Design Round
The same core issue as the previous round of inspection——Destination Recommendation System. This round is more focused on system design, focusing on:
- Multi-model parallel training and serving
- Real-time feature updates
- Large-scale user cold start problem
- System scalability and fault tolerance mechanisms
Lyft’s exclusive sharing of real test questions
Rotting Oranges
This is the most commonly asked mid-level question in interviews. The core test is breadth-first search (BFS), because decay spreads layer by layer, which perfectly matches the layer-order traversal characteristics of BFS.
Topic
Give you one M x n Grid, each grid has three values:
- 0: empty cell
- 1: fresh oranges
- 2: rotten oranges
Every minute, fresh oranges adjacent to each other in four directions will be rotted.
Returns the minimum number of minutes to allow all oranges to rot, or -1 if all oranges cannot be allowed to rot.
Example description
- Enter:
[[2,1,1],[1,1,0],[0,1,1]]→ Output: 4 - Enter:
[[2,1,1],[0,1,1],[1,0,1]]→ Output: -1 (the orange in the lower left corner will never rot) - Enter:
[[0,2]]→ Output: 0 (no fresh oranges)
Problem solving steps
- Initialize queue: Traverse the grid, add all the initial rotten oranges to the queue, and count the total number of fresh oranges at the same time.
- Boundary judgment: If there are no fresh oranges, return 0 directly.
- BFS spreading rot:
- Traverse by layer (each layer = 1 minute)
- Take out all the rotten oranges from the current layer each time and spread them in four directions.
- When encountering a fresh orange, it is marked as rotten. The number of fresh oranges is - 1. Add to the next layer of queue.
- Judgment of results: After the traversal, if the number of fresh oranges = 0 → return time; otherwise, return – 1.
Read N Characters Given read4 II – Call Multiple Times
This is a hard-core difficult question and a high-frequency interview design question. The core difficulty is that the read function will be called multiple times and the remaining characters read from the last time must be saved.
Topic
You have an API function Read4(buf):
- Function: Read up to 4 characters from the file and store
BufArray - Return value: The actual number of characters read (0 will be returned after the file is read)
Requirements: use Read4 Accomplish Read(buf, n) Function:
- Function: Read exactly n characters from the file
- Key: The read function may be called multiple times! (You must remember the remaining characters that were not read last time)
Return value: The total number of characters actually successfully read (the remaining length is returned after reading the file)
Problem-solving ideas
Three "member variables" must be used to save the state (the key to multiple calls)
- Cache array buffer:Save the 4 characters read at once by read4
- Buffer pointer bufferPtr: Points to the next location in the cache to be read.
- Buffer length bufferCount: The actual number of valid characters in the cache
Execution logic
- Eat cache first: If there are remaining characters from the last time, cache will be used first.
- Read new ones: The cache is used up before calling read4 to read the new 4 characters.
- Read and save: Copy characters to the target buf until n characters are read or the file is read.
- Stay status: The remaining characters are stored in the cache for use in the next call.
Exam preparation advice
If you are also preparing for Lyft MLE 2026 (Intern / New Grad / Social Recruitment), please leave a message or private message to communicate:
- Want to see K-Means handwritten code or recommendation system design notes?
- Need a detailed framework for ML System Design?
- Want to learn about Behavioral’s method for preparing high-frequency questions?
Also, if you are pressed for time or want to prepare systematically, it is recommended Programhelp Professional assistance services. They provide OA ghostwriting,VO real-time thinking assistance , full-process coverage and other support, many students successfully obtained Lyft offers through their help.
I wish everyone good luck in your Lyft interview and get your favorite offer as soon as possible!
