Upstart Interview Questions — Page 5

Upstart·Software Engineer·Technical Phone Screen

May 2026

Short coding screen at Upstart for a software engineer role. Just one problem, nothing too wild, but it tripped me up more than I expected.

Given a list of coordinate points, return an array representing the bounding rectangle: the minimum x, minimum y, the total width, and the total height.

“Took me a beat to realize they wanted width and height as differences, not the max values themselves.”

View Post

Upstart·Data Scientist·Technical Phone Screen

May 2026

Second technical round at Upstart for a Data Scientist role, focused entirely on a sliding window coding problem. Pretty standard algorithmic interview but the K distinct characters twist required more careful implementation than I expected.

Given a string and an integer K, write an algorithm that finds the length of the longest substring containing at most K distinct characters.

“I knew it was a sliding window problem pretty fast, which felt good, but then I fumbled the shrink condition for longer than I'd like to admit.”

View Post

Upstart·Software Engineer·Technical Phone Screen

May 2026

Research Scientist interview at Upstart that went deep into probabilistic modeling and simulation. The core problem was meaty enough that it probably took up most of the session on its own.

Build a Monte Carlo simulation to estimate the expected number of players who get banned in a coin-flip game where each player makes up to 100 guesses and gets banned after 10 consecutive correct guesses. Then compare your simulation result to an analytical solution and talk through how many trials you need for stable estimates, plus any variance reduction techniques.

“This one had a lot of layers.”

View Post

Upstart·Data Scientist·Technical Phone Screen

May 2026

Upstart Data Scientist interview with a Python coding question built around clickstream data. Pretty much one meaty technical problem with some complexity analysis tacked on at the end.

Write a Python function that takes a list of event dictionaries (each with user_id, event_type, revenue, and timestamp) and returns per-user total revenue and average session duration in seconds. Walk through the time and space complexity of your solution.

“I went straight for defaultdict and did a single pass, tracking min and max timestamps per user to get session duration.”

View Post

Upstart·Software Engineer·Online Assessment (OA)

May 2026

Took an online assessment for a software engineer role at Upstart. One coding problem, 40 minutes, grid traversal. Did not go well.

Given a 2D grid, find a compromised server by navigating the grid using a function that takes a column, row, direction (one of 8 possible), width, and height. You have a maximum number of attempts.

“My approach moved exactly one step per call and passed two test cases, but the large data cases all timed out.”

View Post

Upstart·Data Scientist·Technical Phone Screen

May 2026

Upstart data scientist interview that leaned heavily into ML theory. The regularization question was basically a mini oral exam and I felt like I was back in a stats class trying to remember things I'd used but never had to articulate out loud.

What are the differences between L1 and L2 regularization in terms of how they modify the objective function, their geometric intuitions, and what they tend to do to model parameters?
How do Ridge and Lasso regression relate to L2 and L1 regularization, and what does each one do to feature selection?
When would you choose Ridge over Lasso, and where does elastic net fit in? What happens to these methods when your features are correlated with each other?
Why does feature scaling matter when applying regularization methods?
Can you define likelihood and explain how it connects to loss functions like negative log-likelihood?

“I knew the mechanics but fumbled the geometry part.”

View Post

Upstart·Product Manager·Onsite - Behavioral / Leadership

May 2026

Second project deep-dive round at Upstart, where they want you to pull out a completely different project from your first one and be ready to defend why you picked it and how it contrasts. Pretty standard for a senior PM loop but the compare-and-contrast angle tripped me up a bit.

Walk me through a second project in depth, covering the problem, your approach, your specific role, key decisions and trade-offs you made, and what the impact was. This should be a different project from the one you already discussed.

“The part that got me was the compare-and-contrast follow-up.”

View Post

Upstart·Software Engineer·Online Assessment (OA)

May 2026

Upstart's software engineer OA had three coding problems that looked reasonable at first glance but the third one had a gotcha that would've burned me if I tried to simulate it naively. Decent set overall, felt more like a competency filter than anything trying to trip you up.

Given a list of 2D integer coordinates, return the bounding box as [minX, minY, width, height], where width and height are the differences between the max and min values on each axis.
Given a list of name-score records and a threshold, filter out scores above the threshold, then return the name with the highest remaining score. Return null if nothing qualifies, and break ties by original order.
On a zero-indexed grid of given width and height, implement a function that moves from a starting cell in one of eight directions for up to maxAttempts steps, stopping early if the boundary would be crossed. Return the final cell. The grid and maxAttempts can be very large, so O(1) is required.

“Pretty mechanical.”

View Post

Upstart·Data Scientist·Technical Phone Screen

May 2026

A technical screen for a Data Scientist role at Upstart that was essentially one long, dense A/B testing case study. The question covered everything from data quality checks to the ship decision, and I spent most of the time wishing I had reviewed variance reduction techniques more recently.

You're given results from an A/B test on a redesigned signup flow with columns for variant, sessions, signups, activation at day 7, p95 latency, support tickets, refund rate, and day-30 revenue. Walk through how you'd validate data quality, including sample ratio mismatch checks, bucketing integrity, and comparing exposure logs to analytics counts.
How would you choose the primary metric and guardrail metrics for this experiment, and justify those choices given the available columns?
How would you compute the treatment effect and confidence intervals for both simple proportion metrics and ratio metrics, and what would you do if the revenue distribution is heavily skewed?
How would you assess whether the experiment ran long enough, and how does that relate to a pre-registered stopping rule and minimum detectable effect?
How would you detect novelty or learning effects in this experiment, and what analyses would you run to separate them from a real treatment effect?
Given that variant B shows better activation but also higher latency and more support tickets, walk through a concrete framework for making the ship or no-ship decision.
Beyond the ship decision, what additional insights would you extract from this experiment data, such as segment-level heterogeneity, funnel step analysis, or form field sensitivity?

“This is where I started strong and then trailed off.”

View Post

Upstart·Software Engineer·Onsite - System Design / Architecture

May 2026

System design round at Upstart for a software engineer role. The whole thing was basically one big question about building a chat system from scratch, which sounds manageable until you realize how many layers they actually want you to cover.

Design a real-time chat application similar to Slack or WhatsApp. Walk through functional requirements, non-functional requirements, APIs, and the high-level architecture.

“This is a lot more surface area than it looks.”

View Post