Tubi Interview Questions — Page 1

Tubi·Machine Learning Engineer·Technical Phone Screen

Apr 2026

Tubitv ML Engineer interview that leaned pretty hard into unsupervised learning fundamentals. The coding portion was a full K-Means implementation from scratch, then they pivoted to GMM and EM algorithm theory right after. Not a ton of fluff, just back-to-back technical questions.

Implement K-Means clustering from scratch given a dataset of shape (n_samples, n_features) and a target number of clusters k. Your solution should handle centroid initialization, point assignment, centroid recomputation, convergence checking, and empty cluster edge cases.
How does Gaussian Mixture Modeling differ from K-Means clustering?
Walk through how you would train a Gaussian Mixture Model using the EM algorithm.
In what scenarios would you choose GMM over K-Means?

“This took longer than I expected.”

View Post

Tubi·Data Scientist·Technical Phone Screen

Apr 2026

Tubi data scientist interview with a pretty deep A/B testing question that covers sample size derivation, sensitivity analysis across like six different scenarios, and overdispersed count data. Felt more like a stats exam than a conversation.

For an A/B test with baseline conversion of 8% and a relative MDE of 8% (so +0.64 percentage points), derive and compute the required per-arm sample size using a normal approximation. Show your z-score terms.
Holding everything else constant from the baseline setup, how does required sample size change qualitatively and quantitatively when: the MDE is halved, alpha tightens to 0.01, power increases to 90%, allocation shifts to 75/25, there is user-level clustering with ICC=0.02 and average cluster size 5, and you add group-sequential monitoring with two equally spaced looks using O'Brien-Fleming boundaries?
If the metric you're testing is overdispersed count data following a negative binomial distribution, how would you re-estimate the required sample size?

“The derivation itself isn't bad if you've done it before.”

View Post

Tubi·Machine Learning Engineer·Technical Phone Screen

May 2026

Concept-check round for an early-career ML engineer at Tubitv covering tree models, training loops, evaluation metrics, embeddings, and transformer basics. Pretty breadth-heavy, less about grinding code and more about whether you can talk through trade-offs like a practitioner.

Explain how decision trees work, then contrast random forests with gradient-boosted trees. Why do ensembles outperform a single tree, and when would you pick gradient boosting over a random forest?
Walk through the full training process for a supervised model: loss function, gradient descent, train/val/test splits, regularization, and how you detect and handle overfitting.
How do you evaluate a model? Why is accuracy often a bad metric, and how do class imbalance and threshold selection change your approach?
What is an embedding, why use it instead of one-hot encoding or raw IDs, how is it learned, and where would you apply one?
Explain the basics of transformers: what self-attention computes, why they replaced RNNs for sequence modeling, and what positional encoding is doing.
For gradient-boosted trees, what does the learning rate control and how does it interact with the number of trees you use?
How would you adapt a train/validation/test split for time-series data where rows aren't independently drawn?
When would you prefer PR-AUC over ROC-AUC and why?
What is the computational complexity of self-attention with respect to sequence length, and why does that matter at scale?

“This is the kind of question where you think you know it until you're mid-sentence.”

View Post

Tubi·Data Scientist·Technical Phone Screen

May 2026

Tubi data science interview threw a deep stats question at me that was way more nuanced than I expected. The whole thing was basically a stress test on whether you actually understand p-values or just memorize the definition.

Explain to a non-technical product manager what a p-value of 0.03 from a two-sided A/B test actually means, and just as importantly, what it does NOT mean. Use a concrete scenario: baseline conversion 5.0%, observed lift of +0.4 percentage points, pooled standard error 0.18 pp. Cover the null hypothesis setup and the 'extremeness under the null' interpretation, why p is not the probability that the null is true, how p-values relate to confidence intervals and statistical power, the dangers of peeking at results early, and finish with both a plain-English explanation you'd give the PM and a Bayesian reframe of the same result.

“This question is a monster.”

View Post

Tubi·Data Scientist·Technical Phone Screen

May 2026

Tubi data scientist interview with a meaty causal inference question that honestly felt more like a take-home problem than a live screen. The question was well-constructed but I spent way too long on the bias directions before getting to the fixes.

Users can opt into seeing a discount banner (treatment), and treated users show higher conversion. Define selection bias, identify at least three mechanisms causing it and the direction of each bias, propose two methods that produce less biased causal estimates along with their assumptions and how you'd test those assumptions, and walk through a small numerical example where the naive estimate is wrong but the corrected method gets you closer to the truth.

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

View Post

Tubi·Machine Learning Engineer·Technical Phone Screen

May 2026

Tubitv ML Engineer interview had a pretty meaty coding portion centered on unsupervised learning. Two connected problems, K-Means from scratch and then extending it into a GMM with EM, which felt like a lot to cover in one session.

Implement K-Means clustering from scratch given an N x D matrix of points and an integer k. Cover centroid initialization, the assignment and update loop, convergence criteria, and write tests to verify correctness.
Extend your K-Means implementation to a Gaussian Mixture Model using the EM algorithm, and explain how GMM differs from K-Means conceptually.

“I went with random initialization first and then talked through k-means++ as the better option.”

View Post

Tubi·Machine Learning Engineer·Onsite - System Design / Architecture

Jun 2026

AI-assisted system design round at Tubitv for an MLE role, focused on building a full movie recommendation pipeline from scratch. You're expected to write actual scaffolding code while reasoning through design decisions, which is a different vibe than a pure whiteboard session.

Design a movie recommendation system for a streaming service end to end, covering data ingestion, feature engineering, model training, evaluation, and low-latency serving.
Walk through the latency budget for your two-stage retrieval and ranking design. What gets precomputed offline versus computed live per request?
How do you construct training negatives from implicit feedback, and what breaks if you just sample them uniformly from the full catalog?
You used an AI tool to generate the training loop. What specific bug or train-serve skew issue would you look for in that generated code, and how would you catch it?
How do you prevent the recommender from getting stuck in a feedback loop, where it keeps surfacing the same content and never explores new or niche titles?
A brand new movie is added to the catalog with zero watch history. Trace exactly how it could show up in someone's recommendations within minutes.
What clarifying questions would you ask before starting this design, and how do the answers change your approach?

“This is basically the whole round in one question.”

View Post

Tubi·Machine Learning Engineer·Technical Phone Screen

Jun 2026

ML breadth interview at Tubitv covering a pretty wide range of fundamentals, from classic regression stuff all the way to transformers and optimizers. They also went applied with a full recommender system walkthrough, which I wasn't totally ready for at that depth.

What causes overfitting, how do you detect it, and what are the main ways to address it?
Explain bagging versus boosting and how each relates to the bias-variance tradeoff.
Walk me through linear regression: the core assumptions, when you'd use closed-form versus gradient descent, and how Ridge, Lasso, and Elastic Net differ.
How does logistic regression work, including the link function, the loss, handling multi-class problems, and model calibration?
Explain how transformers work: self-attention, multi-head attention, positional encoding, and the difference between encoder and decoder architectures.
Compare SGD and Adam as optimizers: how momentum and adaptive learning rates work, when each performs better, and what pitfalls to watch for.
How do you approach hyperparameter tuning in practice, and what's your view on grid search versus random search versus Bayesian optimization?
Walk through a recommender system you've built end to end: candidate generation, ranking, feature engineering, training/serving skew, and how you evaluated it both offline and online.

“Felt solid here.”

View Post

Tubi·Machine Learning Engineer·Technical Phone Screen

Jun 2026

Interviewed for an MLE role at Tubitv and it was a pretty dense technical screen covering everything from basic ML concepts to a full recommendation system design. A lot of ground to cover in one sitting.

What is overfitting and what are the common ways to prevent it?
Explain bagging and describe when it actually helps in practice.
Walk me through linear regression and logistic regression, including their practical differences.
How do transformer models work and what are their key trade-offs?
Compare SGD and Adam as optimizers. When would you prefer one over the other?
How would you approach hyperparameter tuning for a model that's going into production?
Design a recommendation system end to end: problem framing, feature selection, model choice, training, evaluation, and handling cold start and feedback loops.

“Pretty standard starting point.”

View Post

Tubi·Data Scientist·Technical Phone Screen

Jun 2026

Tubi data scientist interview with a single very involved measurement question about Super Bowl ad effectiveness. The whole thing was basically one giant case study broken into seven sub-parts, which felt more like a take-home exam than a conversation.

Your app ran a 30-second national Super Bowl ad. Design a full measurement plan to estimate the incremental impact on installs and revenue, covering KPIs and measurement windows, identification strategies without a clean control group, a difference-in-differences setup and its assumptions, confounders, uncertainty estimation, falsification checks, and a power calculation for detecting a 4% lift in daily installs.

“This was one question with seven sub-parts and I genuinely did not expect the power calculation at the end.”

View Post