Skip to main content
📊

Data Scientist Interview Prep — statistics, ML, product sense — for engineers in Tashkent and Central Asia

Data science interviews probe three layers — statistics, machine learning, and product sense — and you have to be defensible on all three. Practice the questions Tashkent fintechs and remote EU employers actually ask, with real-time AI scoring on communication, technical depth, problem-solving, and cultural fit.

Data scientist roles in Tashkent are concentrated at fintechs (Click, Uzum, Payme), e-commerce (Uzum Market), and a small but growing set of EU remote employers hiring for experimentation platforms. The interview format is consistent: SQL screen, statistics and ML round, a product-sense case study (“our DAU dropped 12% — what do you do?”), and a behavioral round.

The questions below are pulled from real Tashkent interviews. Practice them out loud in the language your real interview will be in.

Core skills tested

  • Statistics: hypothesis testing, p-values, confidence intervals
  • A/B testing: power, MDE, peeking, multiple comparisons
  • Classical ML: regression, trees, gradient boosting
  • Feature engineering and leakage avoidance
  • SQL fluency for analysis (window functions, cohorting)
  • Python data stack: pandas, scikit-learn, numpy
  • Product sense: framing problems, picking metrics
  • Communicating findings to non-technical stakeholders

Salary ranges in Tashkent (2026)

Approximate. Remote-first European roles typically pay 30–50% above local rates.

Junior

10–15M UZS / month

Mid-level

20–30M UZS / month

Senior

38M+ UZS / month (or EUR remote)

What you will actually be asked

Pulled from real interviews recorded on NextSuhbat. Each item is a question you should expect, plus what the interviewer is really testing.

  1. 1

    Recruiter screen

    Walk me through a recent analysis or model that changed a decision.

    Why it is asked: Tests product impact. Lead with the decision that changed, not the model architecture.

  2. 2

    Technical

    Explain p-value to a product manager who has not seen statistics in five years.

    Why it is asked: Communication test. If you say "probability the null hypothesis is true," you fail. Be precise.

  3. 3

    Technical

    You ran an A/B test, p = 0.04, lift = +1.2%. Would you ship?

    Why it is asked: Tests whether you reach for power calculations and effect-size practical significance, not just the p-threshold.

  4. 4

    Technical

    Explain bias-variance tradeoff with one concrete example.

    Why it is asked: Bar for any ML role. The example matters more than the definition.

  5. 5

    Technical

    When would you use logistic regression instead of gradient boosting?

    Why it is asked: Interpretability, sample size, regulatory constraints, deployment cost. Have a real opinion.

  6. 6

    Coding

    Given a churn dataset, walk through the steps you would take from raw data to a model in production.

    Why it is asked: EDA, leakage check, baseline, validation strategy, deployment, monitoring. The flow matters more than the syntax.

  7. 7

    Coding

    Write SQL to compute weekly retention cohorts.

    Why it is asked: Window functions, date math, anti-joins. Real product-analytics task.

  8. 8

    System design

    A product manager says DAU dropped 12% week-over-week. Walk me through how you investigate.

    Why it is asked: Cover segmentation (geography, platform, cohort), instrumentation issues, external events, weekend effect, and when to escalate.

  9. 9

    Behavioral

    Tell me about a model you pushed to production that performed worse than expected.

    Why it is asked: Specific gap between offline and online metrics. What you noticed, how you debugged, what you changed.

  10. 10

    Behavioral

    How do you push back when a stakeholder wants a metric that you do not think is the right one?

    Why it is asked: Tests judgment and communication. Pick a real example. Avoid sounding like you always cave or always fight.

Practice these questions out loud — for free

Reading is not practice. Run a 20-minute AI mock interview in English, Russian, or Uzbek and get a scorecard against communication, technical depth, problem-solving, and cultural fit.

Start free mock interview

Built in Tashkent for Central Asia. All practice sessions support English, Russian, and Uzbek voice.