Skip to content
ML Academy

Week 12 · Bonus — optional extra studio

Bonus studio 6: interview loop dry run - case, system design and behavioural

Project thread forecasting · linked lessons day 58, day 59, day 60

Optional extension of Week 12: a timed interview loop with a case, a system design and two behavioural stories, scored against a rubric, with three dated drills. It uses the synthetic airline data and the learner's own project artefacts. No new claims about real companies.

  • 90 minsession length
  • 4timed blocks
  • 5tasks
  • —status

Session plan90 min

  1. 30′Case interview - data analysis
  2. 30′System design - the forecasting service
  3. 15′Behavioural stories
  4. 15′Debrief and drills

Before you start

Prepared before the session (5 minutes). A timer, a blank page, your three project READMEs and decision logs, and python/artifacts/airline_sample.sqlite with the sqlite3 CLI or a Python session. A human partner is optional. No installs.

Interviewer card (simulated). Read the case aloud, answer clarifying questions from the data below only, and say "I don't know, what would you assume?" for anything else.

Case. "A regional leadership slide says that spring bookings fell on our three Izmir, Antalya and Frankfurt routes (IST-ADB, IST-AYT, IST-FRA) after a competitor added seats, and the plan is to cut fares by 10% on those three routes. Do you agree?" Clarification facts you may give: the period of interest is 1 March to 30 June; the previous year is the comparison; the sample is synthetic; competitor seat capacity on those routes rose by 25% on 1 March 2024; the simulated director cares about revenue per flight and load factor; the fare cut would apply to all flights on the three routes.

Case follow-up. "If you were wrong about the cause, what is the cost?"

System-design prompt. "Design the service that, every morning, forecasts demand 14 days out for the 19 daily flights and recommends whether to close the lowest fare bucket. It must be explainable to analysts, survive a missing snapshot, and not use information from after the decision time." Facts you may give: 19 flights, 11 routes, 3 markets; the booking-curve table has 13 snapshots per departure (90, 60, 45, 30, 21, 14, 10, 7, 5, 3, 2, 1, 0 days out); the analysts check the recommendation by 08:00 local time; the team has one engineer.

Failure scenarios (interviewer picks two). A: a competitor adds capacity on one route. B: the 14-day snapshot is missing for three flights on a day. C: a holiday calendar update changes the flag of a past date. D: a retrain makes the forecast better on average but worse on one route.

Blocks4

  1. Block 1: Case interview - data analysis

    30 min

    Pick the role you are interviewing for (Senior Data Scientist, Senior Data Analyst, Machine Learning Engineer or Applied AI Engineer) and write it at the top of your answer sheet. The interviewer is simulated: it is the card in setup, read aloud by a partner or by you with a timer. Its opinions are not a real person's.

    1. Minutes 0-5 - clarify. Ask your questions before you open any data (task case-clarify). Write the questions and what each answer would change.
    2. Minutes 5-22 - analyse. Query python/artifacts/airline_sample.sqlite and work toward a recommendation (task case-analysis). Think aloud in writing: what you checked and why.
    3. Minutes 22-30 - present. Give your recommendation in under 90 seconds, then answer the card's follow-up (task case-analysis continues). Do not edit the analysis during this step.

  2. Block 2: System design - the forecasting service

    30 min

    Same simulated interviewer. Work from a blank page and use the same data-contract language you used for your projects.

    1. Minutes 0-5 - requirements. Questions, assumptions, scale, and what "good" means (task design-service part 1).
    2. Minutes 5-22 - design. Draw or list the data, training, serving, evaluation and monitoring parts and the decision they support.
    3. Minutes 22-30 - break it. The interviewer picks two failure scenarios from the card. Say how each one would show up, how long until you notice, and what you do.

  3. Block 3: Behavioural stories

    15 min

    Two stories, three minutes each, from your own work in this academy or earlier jobs. Use only events that happened and numbers you can trace to an artefact (task behavioural-stories). Then spend the remaining time answering one probing follow-up per story: "What would you do differently?" and "What did you personally do?"

  4. Block 4: Debrief and drills

    15 min

    Score yourself with the rubric below, one line of evidence per row, and then compare the case numbers with the answer key in case-analysis. Write the three drills for the next two weeks (task debrief-drills): one practice item per weakest rubric row, each with a time limit and a pass condition. Save the transcript or written answers.

Tasks5

Sign in to save your work.

  1. Task 1

    analysis

    Write the clarifying questions you would ask before touching the data, in priority order. For each, one line: what the answer would change in the analysis or the recommendation.

    The expected evidence for this task is shown after you save your own output.

  2. Task 2

    analysis

    Analyse the case with the airline database and give a recommendation in under 90 seconds of speaking time (about 200 words written). Show the number or numbers on which the recommendation rests, say what you checked to rule out a competing explanation, and answer the card's follow-up. Save the queries.

    The expected evidence for this task is shown after you save your own output.

  3. Task 3

    review

    Write the design on one page. It must include the grain and the decision time, the data and features available at that time, the baseline, how it is validated, the serving shape, the failure modes you protect against, how it is monitored, who is accountable for the decision, and the cost of running it. Then answer the two scenarios the interviewer picked.

    The expected evidence for this task is shown after you save your own output.

  4. Task 4

    defense

    Tell two stories, each under three minutes, using only things that happened. One must be about a result you had to narrow, correct or retract. The other must be about a disagreement or a trade-off with another person or with a constraint. For each, name the artefact that shows the numbers.

    The expected evidence for this task is shown after you save your own output.

  5. Task 5

    writing

    Score each rubric row below with one line of evidence, compare your case numbers with the key, and write three drills. Each drill has a time limit, a pass condition and a date within the next two weeks.

    The expected evidence for this task is shown after you save your own output.

Rubric5 criteria

CriterionMeetsExceeds
Case structureClarifies before analysing, states the decision and the comparison, and reaches a recommendation that follows from the numbers, within the time.Names the competing explanation they tested, and what would have changed the recommendation.
Quantitative judgmentThe recommendation rests on a comparison the learner can defend, and the numbers are right to the stated precision.Checks the comparison on a period where nothing happened, and sizes the cost of being wrong.
System designThe design is tied to the decision and the decision time, has a baseline, a validation design that respects time, failure handling and monitoring, and a cost.Gates changes on slices, and says how a bad release would be noticed and reversed.
Behavioural evidenceTwo stories use first-person actions, a result traceable to an artefact and an honest account of an error.The learner volunteers what they would not repeat and the control that now exists.
Self-assessmentThe rubric is scored with evidence per row, the case is compared with the key, and three dated drills target the weakest rows.The drills have pass conditions that can be checked by another person.

Required output

A recorded or written loop: the chosen role, clarifying questions, case queries and recommendation, the one-page service design with the two scenario answers, two behavioural stories with their artefacts, the scored rubric and three dated drills.

Optional extension of Week 12; it adds interview-practice evidence and does not change completion counts.

Studios are tracked separately from the 60 core lessons; completing one never completes a lesson.