Upstart interview notes, page 10 of 10

Mid-sized

Most of these Upstart write-ups come from Software Engineers (51), then Data Scientists (42) and Machine Learning Engineers (2). The most reported rounds are the technical phone screen (73), the behavioral / leadership round (8), the online assessment (4) and the multi-round onsite (4).

97 interview notes · updated Jul 2026

Upstart·Data Scientist·Technical Phone Screen

Jul 2026

Upstart data scientist interview with a machine learning coding exercise focused on regression that generalizes outside the training distribution. Pretty technical and hands-on, less about theory and more about whether you can actually write code and justify your choices.

Build a regression model in code that fits training data well but also extrapolates accurately to test points outside the feature ranges seen during training. Walk through your feature engineering, model selection, regularization approach, and how you'd measure extrapolation performance.

“This tripped me up more than I expected.”

View Post

Upstart·Data Scientist·Technical Phone Screen

Jul 2026

Upstart data scientist interview with a pretty involved R/dplyr coding problem. One question but it had multiple parts stacked on top of each other, so it felt like a full technical screen compressed into a single prompt.

Using R and dplyr with provided transactions, refunds, and users tables: (1) compute net revenue per day and channel over a specified date window, where net revenue accounts for refunds occurring within that window; (2) for each user, find their first purchase date and flag whether that first order was fully refunded by the window end; (3) compute a 3-day rolling sum of net revenue by channel; (4) return two tidy data frames with specific column structures. No loops, idiomatic dplyr only.

“This one took me a while to even parse.”

View Post

Upstart·Software Engineer·Technical Phone Screen

Jul 2026

Coding screen for a software engineer role at Upstart. One question, but it had enough layers to it that I spent more time thinking about edge cases than the actual logic.

Given a vocabulary list and a string of scrambled words, decode the string by matching each scrambled word to the correct vocab entry. Each scrambled word is an anagram of exactly one vocab word, and both share the same first and last character.

“The first/last character constraint is what makes this tractable when vocab has anagram pairs like 'pears' and 'spear'.”

View Post

Upstart·Data Scientist·Technical Phone Screen

Jul 2026

Upstart data science interview with two meaty technical sections: one on causal inference without an experiment, and one on multi-variant testing plus post-launch forecasting. Both required real depth and I left feeling like I'd only half-nailed each one.

You need to measure the causal impact of a product change but cannot run a randomized experiment due to legal, platform, or risk constraints. Walk through an end-to-end observational approach, including what you're estimating, the assumptions you need, the failure modes, and how you'd validate it.
How would you reason about whether the short-term causal estimate you measured reflects long-term impact, and what additional data or modeling would you need?
Given three experiment variants with different visit and purchase counts, which variant is winning? Provide point estimates, quantify uncertainty, and explain how you handle multiple comparisons.
After launching the winning variant to 100% traffic, how would you forecast post-launch conversion rate? What factors could make it diverge from the experiment result, and how would you monitor the forecast?

“This is the kind of question that sounds manageable until you're actually in it.”

View Post

Upstart·Data Scientist·Technical Phone Screen

Jul 2026

Upstart data scientist interview with two technical sections back to back: a multi-table SQL problem and a pair of Python/Pandas questions. Pretty heads-down stuff, no behavioral fluff, just code.

Write a single SQL query to compute click-through rate (CTR) broken out by pin format, but only for US users who signed up within 30 days of the event date. Return impressions, clicks, CTR rounded to 4 decimal places, and distinct new user count per pin format.
Given a DataFrame with pin engagement data and a category ID to name mapping dict, find the category name with the highest average time spent among video pins (case-insensitive match). Break ties by lexicographic order, exclude nulls, and do it without looping over rows.
Given a dict mapping users to lists of pin IDs (with possible duplicates), compute the mean number of unique pins per user. Handle empty lists, very large inputs, and explain time and space complexity.

“This one had a lot of moving parts.”

View Post

Upstart·Data Scientist·Technical Phone Screen

Jul 2026

SQL-heavy technical screen for a Data Scientist role at Upstart. Three prompts all built around the same two-table schema, escalating from aggregation to attribution to a filtered count. The edge cases were the real test, not the queries themselves.

Given a table of marketing touches, compute for each calendar month the average number of touches per company, where the denominator is only companies that had at least one touch in that month (not all companies in the dataset).
For each company that converted, find the last marketing touch at or before the conversion timestamp. Return the company, conversion time, last touch time, channel, campaign, and a flag for whether the touch and conversion happened in the same calendar month. Exclude companies with no prior touches.
Count the distinct companies where the last touch and conversion fall in the same calendar month. As a follow-up, also produce a version that additionally requires the gap between conversion and last touch to be 45 days or fewer.

“The denominator scoping is where people slip up.”

View Post

Upstart·Data Scientist·Technical Phone Screen

Jul 2026

Technical phone screen for a Data Scientist role at Upstart. Two coding problems back to back, one simulation-heavy and one pure math/algorithms. Nothing too wild but the trailing zeros question tripped me up more than I expected.

Write a Python function to simulate radioactive decay with a half-life of one day. The function takes the number of days and number of atoms as inputs and returns the survival state of each atom. Also discuss how you'd verify the result against the theoretical distribution.
Implement a factorial function in Python and describe multiple approaches you could use.
Given an integer n, compute the number of trailing zeros in n! without explicitly calculating the factorial.

“I went with the straightforward approach: for each atom, flip a coin m times and check if it survived all of them.”

View Post