About Honuteki — machine learning education
Scoring models,
explained PROPERLY
Honuteki is a Canadian platform built for people who work with credit data, risk assessment, or analytical pipelines and want to understand the machine learning behind scoring — not just run a library and hope it works.
Where scoring meets genuine understanding
Credit scoring has a technical surface and a statistical core. Many practitioners interact only with the surface — they tune hyperparameters, read AUC scores, and ship models without knowing why logistic regression outperforms gradient boosting on sparse tabular data or when it does not. Honuteki was built to close that gap through structured learning, not slide decks.
Each module runs as an interactive quiz sequence rather than a passive video. A learner encounters a scenario — a dataset with missing bureau features, a model with a Gini coefficient of 0.38, a population shift between training and deployment — and navigates it with guided feedback at every step. Wrong answers produce an explanation, not just a red mark.
Who built this
The curriculum was designed by Tobias Reinfeld, who spent eight years building scorecard systems for a mid-size Canadian lender before moving into education. Adaeze Okonkwo led the learning experience design, drawing on instructional research in STEM contexts. Lukáš Vrábel manages the ML content layer and updates modules when methods shift in the field.
How the platform developed
What makes the content specific
Actual tabular data problems
Modules use datasets structured like real bureau feeds — sparse features, population-level distributions, class imbalance ratios typical of Canadian consumer lending. No toy datasets with perfectly separated classes.
Model choice reasoning
Each assessment presents a problem context and asks the learner to justify a method. Choosing logistic regression requires explaining why interpretability matters here. Choosing XGBoost requires acknowledging the monitoring trade-off.
Regulatory awareness built in
Canadian lending operates under FCAC guidelines and provincial privacy law. Content does not treat regulation as a footnote — it appears alongside the technical material as a constraint that shapes valid model design, not just compliance checkbox.