foundry-data

Plugin foundry-data from marketplace foundry in fedcal/foundry. Commands read from the repo's manifest — 1 month ago.

Skills (4)

• evaluate-model

  Evaluate a trained model the way its problem type actually demands — the right metric family for classification, regression, ranking or forecasting, calibration checked rather than assumed, performance broken out by subgroup, comparison against the baseline restated with uncertainty, error analysis

• explore-dataset

  Reproducible exploratory analysis before any model is trained — detect the language and ecosystem in play (Python/pandas, Java data mining, R, SQL-only), profile shape and types, quantify missingness and outliers, check correlations, hunt target and temporal leakage, and reach an explicit verdict on

• productionise-notebook

  Turn an exploratory notebook or script into reproducible, servable code — extract logic into modules with tests, pin seeds and dependency versions, package it for how it will actually run (batch job, service endpoint, scheduled pipeline, or a Java artifact), and wire the minimum monitoring that catc

• train-model

  Go from a trivial baseline to a defensible model — pick the split strategy correctly (temporal, grouped or random, treated as three genuinely different cases), cross-validate honestly, engineer features without leaking the target, tune within the fold, and pass the model through the single gate that

Subagents (4)

• data-analyst

  Exploratory data analysis as a discipline, not a formality — dataset profiling, data quality and integrity checks, missing-value and outlier characterisation, target-leakage detection, and descriptive plus inferential statistics that report effect size and confidence interval rather than a bare p-va

• ml-engineer

  Classical machine learning model building — feature engineering and encoding, mandatory baseline before any complex model, split and cross-validation strategy chosen for the data's actual structure (i.i.d., grouped or temporal), leakage-safe pipelines, hyperparameter tuning, overfitting detection, a

• mlops-engineer

  Takes a model from notebook or standalone training code to a production system that can be trusted to keep running correctly — data and model versioning, experiment tracking, reproducible training runs, serving pattern selection (batch, real-time, embedded), production monitoring for both system hea

• model-evaluator

  Trained-model evaluation beyond accuracy — metric selection matched to the problem (imbalance, cost asymmetry, ranking vs classification vs regression), confusion-matrix and error analysis broken down per subgroup, probability calibration, bias/fairness measurement across protected or sensitive attr

Similar plugins