The setting

Juniper Finance underwrites small-business loans, about 18,000 active loans averaging 28,000 dollars, with a portfolio default rate near 12 percent. You are a data scientist on the credit risk modeling team.

Every mission on this path happens at the same company, so context carries over the way it does in a real job: the data you cleaned in mission two is the data the finance lead questions in mission four.

The missions

1. problem framingstarter

Frame the default model request from Juniper's credit risk lead

Alicia Trent, head of credit risk at Juniper Finance, sent a two-paragraph request asking for a model to flag risky borrowers. It is vague in the way real requests are vague, and if you build to it as written you will build the wrong thing.

You deliver: A problem framing memo defining the target, label, horizon, metric, decision use, and known risks.

Scored on: Clear target and label, Metric fits the problem, Catches label leakage, Scope is separated.

Working from: credit_model_request.md, data_availability.md.

2. feature designstarter

Design the feature set for Juniper's default model

With the problem framed, senior data scientist Ravi Menon wants a feature design before anyone writes training code. Juniper sends adverse action notices, so every feature has to be explainable to an applicant who was declined.

You deliver: A feature design doc listing proposed features, transformations, and an excluded list with reasons.

Scored on: Application-time features only, Useful engineered features, Excludes leaky columns, Flags proxy risks.

Working from: loan_data_dictionary.md, feature_notes.md.

3. model evaluationcore

Judge model_a against model_b before Juniper promotes one

Two candidate models are on the table. A product manager is pushing to promote model_b because it has a higher AUC and higher accuracy. Before Juniper swaps its champion you need to say which model actually serves the lending decision.

You deliver: A model evaluation report recommending one model with a quantified trade-off and a critique of the accuracy argument.

Scored on: Reads the operating point, Quantifies the trade-off, Critiques accuracy and AUC, Clear recommendation.

Working from: model_scorecard.csv, confusion_matrix.csv, eval_context.md.

4. bias checkscore

Run the fairness slice review on Juniper's default model

Before the model goes to compliance, Juniper requires a fairness review across borrower segments. The headline metrics look fine, but headline metrics hide subgroup harm.

You deliver: A fairness assessment naming the harmed subgroup, quantifying the disparity, and recommending mitigations.

Scored on: Finds the harmed subgroup, Quantifies the disparity, Applies the parity guideline, Concrete mitigations.

Working from: fairness_slices.csv, fairness_policy.md.

5. deployment handoffstretch

Write the deployment handoff so the next null-revenue incident gets caught

The model is approved and going to production. Last time Juniper shipped a model, a pipeline change silently set annual_revenue to null for a chunk of applications and nobody noticed for three weeks while approvals drifted and defaults climbed. That cannot repeat.

You deliver: A deployment handoff covering monitoring signals, alert thresholds, rollback, retraining triggers, and ownership.

Scored on: Input monitoring, Concrete alert thresholds, Rollback and retraining, Would catch the prior incident.

Working from: model_card_draft.md, serving_notes.md, prior_incident.md.

How the scoring works

Each deliverable is graded against the rubric written for that mission. Separately, every mission on every path is graded on how you used AI, against the same four criteria:

  • Understood the task. The learner framed the goal for the assistant clearly instead of pasting the brief and hoping.
  • Grounded in the material. The learner directed the assistant into the provided files and based the work on them, not on invented facts.
  • Verified the output. The learner checked claims, numbers, or coverage against the source material before submitting.
  • Iterated with judgment. The learner refined weak parts of the draft with specific follow-ups rather than accepting the first answer.

Both scores, with the work behind them, go on your proof profile. That is what makes a claim like "I can use AI for data science and ml" something an employer can check.