这场 Uber 's Virtual Onsite (Coding session) was one of the most impressive I've led in recent memory.
45 minutes, 3 questions ranging from Algorithm basics to SQL & Pandas to cohort analysis, with almost no wasted topics. For anyone looking to break into Uber DS / Scientist / DE positions, this set of questions is worth breaking down carefully.
Question 1: Implement sqrt() without using any external library
Requirements: implement a function to calculate the square root without using any library function sqrt(x).
The trainee was stuck here, and wrote a straightforward 0-to-x/2 grid searchThe interviewer said, "Although the logic is fine, the complexity is O(n), and it can't run with a large number of inputs. The interviewer didn't interrupt right away, but the expression on his face already said it all.
The correct thinking should be:
- Use binary search Do a bisection lookup.
- each time you take
mid = (left + right) / 2judgementmid * midwith respect to x, and keep shrinking the interval until the error is small enough. - If decimals are allowed, set the precision to 1e-6.
It's not really about implementing square roots, it's about constructing your own algorithmic ideas, and Uber loves to come up with handwritten "no library functions allowed" questions like log(), pow(), sqrt(), and abs(), which rely on the algorithmic logic in your head.
Question 2: Identify the top three drivers in each city who made the most money last month
The table structure is as follows:
(driver_id, date, trip_id, payment, city_id, is_completed)
Ask to find the "top three drivers who made the most money in each city last month" using SQL or pandas.
This question has a strong Uber flavor at first glance - it's very business oriented, but it's really about your proficiency in data grouping + ranking functions.
SQL Writeup:
WITH monthly AS (
SELECT
city_id, driver_id
city_id, driver_id, SUM(payment) AS total_payment
SUM(payment) AS total_payment
FROM trips
WHERE date >= DATE_TRUNC('month', CURRENT_DATE - interval '1 month')
AND date < DATE_TRUNC('month', CURRENT_DATE)
AND is_completed = TRUE
GROUP BY city_id, driver_id
)
SELECT *
FROM (
SELECT
city_id, driver_id, total_payment, ROW_NUMBER(
ROW_NUMBER() OVER(PARTITION BY city_id ORDER BY total_payment DESC) AS rn
FROM monthly
) t
WHERE rn <= 3.
Interviewer focus look:
- You are interested in the window function (
ROW_NUMBER,RANK,DENSE_RANK) Proficiency or lack thereof - Flexibility to reproduce SQL logic in Pandas
- Whether filtering conditions have been considered (e.g.
is_completed = TRUE)
Question 3: Of the drivers who signup each month, identify the top three highest earners in each city.
Add a new table:
(driver_id, signup_date)
Require:
Find the top three drivers who make the most money in each city among the drivers who sign-up each month.
This question is an advanced version of the second question, with the additional cohort dimension of "signup_month", which Uber likes to test, especially in Scientist interviews.
SQL Solution:
WITH cohort AS (
SELECT
s.driver_id,
DATE_TRUNC('month', s.signup_date) AS signup_month,
d.city_id, SUM(d.payment) AS total payment
SUM(d.payment) AS total_payment
FROM trips d
JOIN signup s
ON d.driver_id = s.driver_id
WHERE d.is_completed = TRUE
GROUP BY 1,2,3
),
ranked AS (
SELECT
*SELECT
ROW_NUMBER() OVER(PARTITION BY signup_month, city_id ORDER BY total_payment DESC) AS rn
FROM cohort
)
SELECT *
FROM ranked
WHERE rn <= 3;
This question is at the heart of the test:
- Can you understand the cohort logic of "registered driver of the month"?
- Can you combine two tables correctly (join + grouping + window functions)
- There's no business sense, like filtering out unfinished orders.
Many candidates get stuck here after writing the second question, in fact the logic is the same as long as you aggregate and then group and sort.
Test Logic for Uber Interviews
In those 45 minutes, Uber is actually testing three competencies:
- algorithmic thinking -- Math implementations that do not rely on library functions
- SQL / Pandas Practical Skills -- Ability to handle large-scale transaction data
- Cohort Analytical Awareness -- cohort performance with enrollment and order data
The overall difficulty level is medium to high, but the proficiency level is extremely high. If you haven't brushed up on similar questions in advance, it will be difficult to write them all in 45 minutes.
Programhelp Guaranteeing your success
We were remotely assisted throughout this VO, especially in the second and third question sections.
- When the trainees began to write in pandas sorting logic mixed up, we voice reminds the groupby + rank usage, immediately pull back the rhythm;
- In the cohort section of question 3, we hinted at "to use signup_month + city_id double dimensional partitioning" to avoid rank errors.
Finally managed to pass all three questions.
Programhelp's remote voice-assisted service specializes in Uber / DoorDash / Lyft / Amazon / Instacart OA + VO hands-on chaperone for data and algorithmic posts such as these.
Through real-time voice prompts and seamless online assistance, we can help you stabilize your output at critical moments, so that you can go from "knowing how to write" to "really being able to write at the interview site".
