← Back to examples
Static GitHub Pages mock End-user preview Fast / Thinking mode No setup required

HUGIML LLM Chat Workbench

A simple preview of the HUGIML chat experience. Choose a dataset, ask questions in plain English, and switch between Fast answers or more polished Thinking explanations.

Chat

Ask the next question credit_risk_synthetic
Fast mode is best for quick answers and dashboard-style facts.

Data profile

Rows6,000
Predictors18
Targetdefault
Positive class28.4%

Predictor groups

Summary statistics

Model evidence

Primary metricROC AUC 0.842
Accuracy0.781
Precision0.736
Recall0.691

Feature / pattern influence

Confusion matrix

Pred 0Pred 1
Actual 0740100
Actual 1111279

Precision is shown together with the confusion matrix so reviewers can see both the headline score and the underlying counts.

Pattern details and breakup

Input features Discovered patterns Added signals Strength and direction

Pruning and governance

Previous result

Patterns184
ROC AUC0.842

Before pruning, the model keeps every discovered pattern.

New result after pruning

Patterns141
ROC AUC0.839

After pruning, weak patterns are removed and the before/after comparison stays visible.

Audit trail

Action: Pruned weak patterns Reason: Make the model easier to review before sign-off Removed: Very rare patterns Rebuilt model: Yes Outputs: Model card, audit summary, pruning log

What happens when you ask a question

1. Ask naturally

Type questions like “What is the precision?”, “Which inputs matter most?”, or “Can you prune weak patterns?” The app turns them into safe HUGIML actions behind the scenes.

Dataset summaryModel qualityKey driversPattern review

2. Choose the answer style

Fast is best when you want the answer quickly. Thinking is best when you want a clearer explanation for a report, review, or stakeholder discussion.

Fast factsPlain-English explanationSame numbers

3. Review the evidence

The chat answer is paired with tables, cards, and pattern details so users can check the result instead of relying on a black-box response.

MetricsFeature importancePattern detailsGovernance summary