CourseAI / ML Engineer: The Complete Course · Module 14: Responsible ML · part 83 of 93
Part 83 · Module 14: Responsible ML

Lesson 1: Fair and explained: who the model fails, and why

29 min read·
Module 14 · Lesson 1 · About 55 minutes · Needs: model v2 and error analysis (Module 6 lesson 2), the warning threshold and costs (Module 3 lesson 3), and the Kestrel data (course-guide/06).

The situation

Week 14 at Kestrel. Hannah forwards two emails: a customer on an island route asks why she is warned so often, and a local councillor asks whether the model treats island customers worse. Adaeze wants both answered with numbers.

One parcel runs through the lesson: KP0088193 (simulated), on island route INV-R09, handed in on Monday 24 August 2026 at 18:00 for next-day delivery, with a new driver. Model v2 gives it 0.231, above the 0.129 threshold, so its customer was warned. It was late.

WHERE THIS FITS The Kestrel system this lesson SHARED SERVICES Depot scans parcels, routes Ingestion nightly batch Validation Pandera checks Data lake Parquet, DuckDB Experiments MLflow runs Training LightGBM, Optuna Online store Feast, Redis Features point in time Gates vs the champion Registry staged versions Serving API FastAPI, ONNX Consumers vans, customers Retraining trigger, budget Alerting on call Monitoring drift, Evidently Late labels a day later Orchestrator Prefect flows Versions DVC, Git CI/CD GitHub Actions Containers Docker images
This lesson checks the trained model (experiments) before the gates and the customer warnings (consumers): blue marks the parts it touches, grey the rest.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.

Buy this course · ₹399See what is inside

Or open everything with a plan