Skip to content
ML Academy

Week 12 · Sunday — consolidate and defend

Capstone hiring-committee simulation (Day 60 extended assessment)

Project thread rag · linked lessons day 56, day 57, day 58, day 59, day 60

Capstone close for all three project threads (forecasting, neural, retrieval): a simulated hiring-committee defense using the learner's own artefacts, an unaided cross-sector transfer answer, a revised answer under challenge, a supported competency report and a 30/60/90-day plan. This is the extended part of the Day 60 final assessment.

  • 90 minsession length
  • 4timed blocks
  • 6tasks
  • —status

Session plan90 min

  1. 20′Spaced recall
  2. 40′Hiring-committee simulation
  3. 20′Teach-back and decision defense
  4. 10′Weekly review and planning

Before you start

Before the session: the three project READMEs and reports, the Week 12 Saturday capstone evidence list, the decision and assumption logs, the competency report generated from recorded evidence, and a timer. A human partner is optional; without one, the app supplies the simulated committee prompts, or you read them from a prepared card and record your answers. No new data, code or installation.

Blocks4

  1. Block 1: Spaced recall

    20 min

    Cumulative, closed notes, mixing Parts I and II; then correct:

    • Part I: the chain grain -> availability -> leakage -> split -> metric -> error slice -> decision, applied to your forecasting project in five lines; one way your backtest could still have been optimistic.
    • Weeks 7-8: what your neural comparison proved and did not prove, in two sentences.
    • Weeks 9-11: recall@5 versus MRR@10, why the held-out part had to stay closed, and the control that stopped the injected chunk in your incident.
    • Day 58: the elements of a decision record and when one should be revisited.
    • Two items from your review queue that you have missed more than once.

  2. Block 2: Hiring-committee simulation

    40 min

    Choose one target role before starting: Senior Data Scientist, Senior Data Analyst, Machine Learning Engineer, or Applied AI Engineer. The committee is simulated: its members (a hiring manager, a technical interviewer and a skeptical peer) are role-play prompts delivered by the app or a human partner; their questions and opinions are not real people's views.

    1. Portfolio walk-through (12 min). About four minutes per project: the decision it supported, the baseline, the key result, one diagnosed failure, and one limitation. Use only artefacts from your portfolio.
    2. Role-specific follow-ups (13 min). The committee asks three follow-ups drawn from your own artefacts and the chosen role (for example: for Data Analyst, metric definitions and stakeholder communication; for ML Engineer, reproduction, monitoring and rollback; for Applied AI, evaluation, injection controls and cost). Answer each in under three minutes.
    3. Transfer case (15 min, unaided). Answer the transfer task below.

  3. Block 3: Teach-back and decision defense

    20 min

    The skeptical peer challenges your weakest answer from the simulation (chosen by the partner or the app, or by you if working alone: the answer you were least able to support with evidence). Take two minutes to revise it with evidence from your artefacts, then answer one further challenge on the same point. Afterwards, write 150 words: what changed between the first and revised answers, and what evidence was missing.

  4. Block 4: Weekly review and planning

    10 min

    Review the supported competency report: every competency marked demonstrated must link to an artefact or assessed answer; anything without evidence is marked not yet demonstrated or unassessed (for example, generated-answer quality when you worked in retrieval-evaluation mode). Then write a 30/60/90-day plan for the chosen role: one concrete practice goal and one portfolio improvement per period, each tied to a gap in the report.

Tasks6

Sign in to save your work.

  1. Task 1

    recall

    Answer the mixed Part I and Part II recall set, including the forecasting chain applied to your own project, and correct misses.

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

  2. Task 2

    defense

    Present the three projects to the simulated committee: decision, baseline, key result, diagnosed failure and limitation for each.

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

  3. Task 3

    defense

    Answer three role-specific follow-ups drawn from your artefacts and the chosen role.

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

  4. Task 4

    analysis

    Answer the five parts of the transfer_task in 15 minutes, without notes and without the reference solution. Save your answer as it stands when the time ends.

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

  5. Task 5

    defense

    Revise your weakest answer under challenge and write what changed and what evidence was missing.

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

  6. Task 6

    writing

    Review the supported competency report for unsupported claims and write a 30/60/90-day plan for the chosen role.

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

Transfer task

New cross-sector case (synthetic; not a real organization). A regional hospital network wants to reduce outpatient no-shows. It has three years of appointment records (booking date, appointment date, department, reminder sent yes/no, prior no-shows, travel distance band) and a folder of about 200 internal policy documents about scheduling, reminders and patient communication. A manager asks for "an AI that predicts no-shows and answers staff questions about the policies".

In 15 minutes and without notes, answer:

  1. Which decisions would each part support, and is a model needed for each? Name the simplest baseline for the prediction part and for the question-answering part.
  2. For the prediction part: the grain, the cutoff (what is known when the decision is made), one leakage risk, and the split you would use.
  3. For the question-answering part: how you would build the first evaluation set, one security or privacy risk specific to health data, and the enforceable control for it.
  4. Which part of your own forecasting, neural or retrieval project transfers directly, which does not, and why.
  5. One cost of being wrong for each part, and the evidence that would make you stop or change the plan.

Rubric5 criteria

CriterionMeetsExceeds
Evidence-based defenseClaims about all three projects are supported by named artefacts; baselines, results, failures and limitations are stated accurately.Volunteers the weakest result and explains what was learned from it without being prompted.
Role fitFollow-up answers address the concerns of the chosen role with a position, evidence and a trade-off.Connects answers across projects (for example, the same leakage or contamination discipline in forecasting and retrieval).
Transfer reasoningThe new case gets a clear framing, a baseline, a validation design, a failure risk and a decision rule adapted to its sector.Identifies which part of the learner's own projects does not transfer and why.
Revision under challengeThe revised answer is better supported than the first and the reflection names the missing evidence.Concedes a valid point and narrows the claim instead of defending it unchanged.
Competency report and planNo competency is marked demonstrated without linked evidence; unassessed areas are explicit; the plan addresses reported gaps.The plan's goals are measurable and include the evidence that would show each gap closed.

Required output

Final project evidence, portfolio review, supported competency report, and 30/60/90-day plan: the recorded hiring-committee simulation for one target role, answers to three role-specific follow-ups, the unaided cross-sector transfer answer, a revised answer under challenge with reflection, the reviewed competency report and the plan.

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