HAXplorer supports analysts exploring large tabular datasets in which both the rows (items) and the columns (attributes) can be meaningfully grouped into multi-level hierarchies. The workflow begins with the analyst configuring how items and attributes should be aggregated — combining semantic criteria drawn from auxiliary metadata (e.g., grouping municipalities by canton, or chess openings by category) with statistical criteria derived from agglomerative clustering of the data values themselves. The system then mechanically builds the resulting item and attribute hierarchies, computes representative aggregate values at every item-aggregate × attribute-aggregate crosscut, and projects items into a two-dimensional similarity space via a dimensionality-reduction technique chosen by the analyst (PCA, t-SNE, or UMAP). Four coordinated views are rendered: an Item Similarity View (projection scatterplot), an Item Hierarchy View and an Attribute Hierarchy View (indented trees supporting drill-down and roll-up), and an Item Attribute Matrix View (a colour-encoded heatmap of the aggregate crosscut values). The analyst then enters an iterative exploration loop: navigating, sorting, filtering, and selecting across the linked views to compare items, attributes, and their aggregates and to detect outliers or recurring patterns, with findings accumulating across iterations. After each round the analyst judges whether enough insight has been gathered; if not, the hierarchy configuration is revised — by adding or removing aggregation layers, changing the clustering resolution or dimensionality-reduction technique, or filtering out distracting items or attributes — the hierarchies and aggregates are recomputed, and all views are re-rendered. The loop repeats until the analyst is satisfied, at which point the accumulated comparisons and outlier findings are synthesised into a final, domain-specific insight (e.g., explaining a canton's outlying voting behaviour, characterising an edition's annotation practice, or describing a historical shift in chess opening preferences).