model-evaluation

Patterns for testing a data pipeline — unit tests for transforms, data-integrity assertions, data-leakage detection, and reproducibility checks.

Use when reading, loading, or connecting to any data source — local files (CSV, TSV, Parquet, JSON, JSONL, Excel), SQL databases, REST APIs, or cloud buckets…

Use to generate standard descriptive statistics for a DataFrame — dtypes, nulls, distributions, cardinality, correlations, and an overall quality score.

Catalog of ready-to-use feature transformations — categorical encoding, numeric transforms, date features, interactions, and selection.

Patterns for using Claude inside the analysis pipeline — interpreting statistical results in natural language, classifying free text, detecting semantic…

Standard templates for executive, technical, and metrics reports.

Use to find academic literature, industry benchmarks, similar public datasets, or prior technical solutions for a data problem.

Conventions and a scaffold for prototyping interactive Streamlit dashboards from pipeline outputs (cleaned data, EDA, modeling, and validation reports).

Use whenever creating data visualizations — histograms, scatter plots, heatmaps, time series, correlation matrices, distributions, or any chart for EDA or…

Approved conventions and patterns for pandas, numpy, scikit-learn, matplotlib, and seaborn in this project.

Rules for handling sensitive data and PII, transformation lineage, and basic compliance.

Patterns for processing large datasets (>1M rows or >500MB) with Dask or Polars.