Twitch Interview Questions & Experiences

Twitch·Data Scientist·Technical Phone Screen

Jul 2026

Twitch data scientist interview that was basically a full SQL gauntlet built around streaming telemetry data. Five tasks back to back, ranging from monthly aggregations to window functions to a conceptual question about query processing order. Felt more like a take-home than a live screen.

Given minute-level streaming data, compute total monthly hours streamed across all streamers. Output should be labeled as YYYY-MM and ordered chronologically, and must handle data spanning multiple years correctly.
For each streamer, compute their total streamed minutes and the share of minutes spent in categories matching a given keyword (case-insensitive substring match). Return the keyword share as a ratio.
For each streamer and calendar month, determine whether their streamed hours increased compared to the immediately preceding calendar month. Treat months with no activity as zero. Only return rows where current month exceeds the previous month.
For 2019 only, return each streamer's average concurrent viewers and total US viewer-minutes. Join minute_viewed to minute_streamed on both streamer and timestamp to avoid row multiplication. As a follow-up, also count distinct streamed minutes where at least one US viewer was present.
Explain SQL's logical query processing order and clarify whether AVG() is computed before or after the JOIN.

“The year-boundary thing is what trips people up.”

View Post

Twitch·Data Scientist·Onsite - System Design / Architecture

Jul 2026

Twitch data scientist system design round, focused entirely on building a real-time recommendation system for live streams. It was a deep, multi-part question and I felt like I was sprinting the whole time.

How would you define the objective and construct training labels for a real-time live stream recommendation system, including handling delayed outcomes and streams that end mid-session?
What features would you use across user, creator, stream, and context dimensions, and how do you handle cold-start for brand new creators or users?
Walk through your model architecture choice, loss function, negative sampling strategy, and how you'd deal with severe class imbalance in this setting.
How would you detect and mitigate position bias and popularity bias in your training data, and what exploration policy would you use?
What offline and online metrics would you track, and how would you validate that offline metric improvements actually translate to online gains?
Describe the end-to-end serving architecture for this system with a sub-100ms p95 latency budget, including how you'd keep embeddings and stream availability signals fresh in near-real time.

“This is where I fumbled a bit at the start.”

View Post

Twitch·Data Scientist·Technical Phone Screen

Jul 2026

Twitch data scientist interview with a heavy SQL focus. Four problems, all streaming analytics, all nastier than they look on paper. The schema is clean but the edge cases are where they actually test you.

Given a streaming platform schema with view start/stop events, write a SQL query to find peak concurrent viewers per stream and the time window where that peak occurs. Events may arrive out of order and exact duplicate rows must be deduplicated.
For users who watched multiple different streams simultaneously on a given date, compute the total minutes each user spent watching two streams concurrently, but only count overlap windows of at least 5 minutes.
Compute D+7 retention for users whose first-ever view was in a specified date range. A user is retained if they watch any stream for at least 2 continuous minutes on exactly their 7th day after first view.
Rank creators by average per-viewer watch time over a 7-day window, restricted to US viewers. Only include creators with at least 100 unique US viewers. Break ties using peak concurrency from the earlier query.

“This one wrecked me more than I expected.”

View Post

Twitch·Data Scientist·Technical Phone Screen

Jun 2026

SQL-heavy technical screen for a Data Scientist role at Twitch, all revolving around two streaming tables. Four questions, each building on the last, and the multi-year edge case stuff caught me more off guard than I'd like to admit.

Using the minute_streamed table, write a query that returns total hours streamed for each calendar month, ordered chronologically. Also explain how your query handles data that spans multiple years.
For each streamer, compute their total streamed hours and the percentage of those hours spent streaming content in a category that matches a given keyword. The match should be case-insensitive.
Identify all streamers whose total streamed hours in a given month exceeded their total from the prior month. Your solution needs to handle multiple years of data and months where a streamer had no activity at all.
Join minute_streamed and minute_viewed to produce, per streamer, their average concurrent viewers in 2019 and the total minutes watched by viewers from the US in 2019.

“Went straight to COUNT(*)/60 grouped by month and forgot to include year in the group-by key.”

View Post

Twitch·Product Manager·Hiring Manager Screen

Jun 2026

Short PM screen at Twitch, basically just one question about motivation for the role.

Why do you want to work at Twitch?

“Pretty standard opener but I fumbled it a bit.”

View Post

Twitch·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

Behavioral round at Twitch for a software engineer role, basically one meaty question about a hard problem you solved. Pretty standard format but the depth they wanted caught me off guard.

Walk me through a time you faced a genuinely difficult problem at work. What made it hard, what did you try (including things that failed), what happened, and what did you take away from it?

“I structured it as situation, then actions, then result, which sounds clean but I spent way too long on the setup and barely had time to get into the stuff that actually mattered: the failed attempts.”

View Post

Twitch·Product Manager·Onsite - Product Sense / Strategy

Jun 2026

PM interview at Twitch with what I can only describe as a market sizing question that made me question everything I know about weddings.

Estimate how many weddings the average American attends over the course of their lifetime.

“Blanked for a second because I kept second-guessing my assumptions about average age ranges and how wedding attendance clusters in your 20s and 30s.”

View Post

Twitch·Product Manager·Onsite - Product Sense / Strategy

Jun 2026

Interviewed for a PM role at Twitch and got asked a pretty technical product question about their API. Short interview, not a lot of fluff.

Give an example of a specific endpoint on Twitch's platform.

“Blanked for a second.”

View Post

Twitch·Software Engineer·Technical Phone Screen

May 2026

Twitch coding screen for a software engineer role, one question about billing logic for a tiered notification service. Pretty self-contained but the output requirements had a small twist that slowed me down.

A notification service bills users based on hourly usage over a 24-hour period. The first N notifications per hour are free; anything beyond that is charged at a fixed per-notification rate. Given an array of 24 hourly notification counts, return both the total daily cost and an array of 24 per-hour charges.

“The core math is simple enough but I tripped up on the output format at first.”

View Post

Twitch·Data Scientist·Technical Phone Screen

May 2026

Twitch data scientist interview that was basically one giant A/B testing case study broken into six parts. It covered everything from randomization unit choice to variance reduction to integrity checks. Dense and technical, felt more like a take-home exam than a conversation.

For an A/B test on reducing pre-roll ad frequency in mobile live streams, how would you choose the randomization unit and why? How do you handle viewers switching between streams and creators sharing audiences, and how do you ensure exposure stickiness across multiple days?
What single primary metric would you choose for this ad frequency experiment, and what guardrail metrics would you include? How do you handle heavy-tailed distributions in watch time data and what does that choice mean for your inference?
Calculate the required per-variant sample size in viewer-days for this experiment. Baseline mean daily watch time is 36 minutes with a standard deviation of 60 minutes, design effect of 1.3, 5 million eligible daily mobile viewers, two-sided alpha of 0.05, 80% power, and a target of detecting a 2% relative lift.
How would you use pre-experiment data to reduce variance in this experiment? Describe the regression you would fit and how you would apply the resulting coefficient to adjust your outcome metric.
Design a two-week ramp-up plan for this experiment that controls type I error under sequential monitoring. What stopping rules or decision boundaries would you set at interim checks, and how do you account for multiple looks at the data?
How would you detect and filter out bot traffic and AFK viewers to protect experiment integrity? How would you handle creator-led raids as a source of bias, and what would you do if you detected spillover effects between variants?

“This tripped me up more than I expected.”

View Post