Can I export visualizations for research papers?

Yes, it generates both interactive HTML files for web viewing and static PNG images suitable for inclusion in documents and presentations.

What visualization types are supported?

The skill supports heatmaps, radar charts, model comparison plots, task-specific visualizations, and factor-based comparisons based on your experiment variables.

What are the prerequisites for running this skill?

You need completed evaluation logs (.eval files), the inspect-viz library installed, and an experiment_summary.yaml file from the design-experiment step.

How does it handle compute resource tracking?

If Slurm logs are available, it extracts job IDs to generate a compute utilization report, including GPU metrics and execution efficiency.

Do I need to manually configure the plots?

No, the skill automatically infers the most appropriate visualization types by reading your experiment_summary.yaml file and detecting available metrics.

LLM Experiment Visualization & Analysis

Name: LLM Experiment Visualization & Analysis
Author: niznik-dev

byniznik-dev

•

Data Science & ML

Generates interactive visualizations and comprehensive evaluation reports from LLM social research experiments.

This skill automates the post-experiment workflow for LLM research by transforming raw evaluation logs into actionable insights. It intelligently parses experimental variables to select the best visualization types—such as model-to-task heatmaps, radar charts for multi-metric analysis, or binary factor comparisons—and produces both interactive HTML plots and static PNGs. Beyond visualization, it generates a detailed Markdown report featuring Wilson score confidence intervals, compute utilization metrics from Slurm logs, and automated interpretation of results to suggest future research directions.

Key Features

01Comprehensive Markdown reports with metrics, 95% confidence intervals, and research interpretations.

02Automated visualization selection including heatmaps, radar charts, and model-task matrices.

03Compute utilization tracking and GPU metric summarization from Slurm logs.

0411 GitHub stars

05Smart data deduplication and metadata extraction from evaluation logs for clean analysis.

06Interactive HTML and static PNG plot generation via inspect-viz and Playwright.

Use Cases

01Auditing compute efficiency and resource usage for large-scale evaluation pipelines.

02Visualizing the impact of specific experimental factors on model accuracy and calibration.

03Comparing performance across multiple LLMs for complex social research tasks.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add niznik-dev/cruijff_kit analyze-experiment

For use in Claude.ai and ChatGPT

Download Skill