Банк вопросов из реальных собеседований

What backend components are needed for a LLM product with tools, cache and long tasks?

A speech product collects user audio. How would you filter and route audio snippets for ASR/TTS training data without poisoning the dataset?

Several annotators transcribe the same low-resource-language audio and produce inconsistent text, numbers and addresses. How do you clean or aggregate the labels?

There are 100 dogs and 100 cats. 50 dogs and 40 cats weigh more than 5 kg. Let B - the animal is a cat, A - the animal weighs more than 5 kg. Find P(A | B).

A fair coin is tossed 10 times. What is the probability of getting exactly 6 heads?

Why do they use BatchNorm? How does it work during training and how does the behavior differ on inferencing?

Explain BatchNorm for image tensors, what changes at inference, and what must be synchronized in multi-GPU training.

How to solve a probability problem with a large number of independent identical tests?

How are PDF, CDF, fitting distributions and Bayes' theorem related?

A coin has unknown heads probability p. You observe n tosses and k heads. Derive the maximum likelihood estimate for p.

Where is temperature applied, how does it change the distribution after softmax, and how is it different from token sampling?

Why does the high accuracy of a medical test not guarantee a high probability of illness after a positive result?

How to calculate the probability of at least one head in 10 tosses of a fair coin?

What is a vector space? What is the difference between a linear span and a basis?

What is a probability space? What is a set of measure zero and why can a finite or countable set have probability zero in a continuous distribution?

Explain the law of large numbers and the central limit theorem. What distribution do you get if you repeatedly take samples of 100 observations and compute their mean?

In a series of coin tosses, the eagle fell H times and the tails T times. How can the maximum probability method be used to estimate the probability of eagle falling p and check that the maximum is found?

How can an analytical solution of linear regression be obtained by the method of least squares, when it is not applicable and under what assumptions the method of maximum probability gives the same result?

How to explain the role of the A/B test in checking for changes in the recommendation system?

What are the two main stages of a recommendation system? Explain the search for candidates and ranking briefly but meaningfully.

How important is it for a data scientist to understand the business logic of how users got into the application? Are just events from the database enough?

Explain what a convolutional neural network is to senior engineers who do not specialize in ML. Keep it accurate but accessible.

OLX needs a single API recommendation for cars, vacancies, real estate and other display points. How to define the interface and boundaries of team responsibility?

How to organize code review in a backend team in order to maintain quality and at the same time increase the team’s knowledge of the code base?