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
Data profile
Predictor groups
Summary statistics
Model evidence
Feature / pattern influence
Confusion matrix
| Pred 0 | Pred 1 | |
|---|---|---|
| Actual 0 | 740 | 100 |
| Actual 1 | 111 | 279 |
Precision is shown together with the confusion matrix so reviewers can see both the headline score and the underlying counts.
Pattern details and breakup
Pruning and governance
Previous result
Before pruning, the model keeps every discovered pattern.
New result after pruning
After pruning, weak patterns are removed and the before/after comparison stays visible.
Audit trail
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.
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.
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.