Pivot Tables in Custom Dashboards
Steffen Schmitz
Analyze multi-dimensional data with the new pivot table widget in Langfuse custom dashboards.
We’re excited to introduce pivot tables to Langfuse custom dashboards! This new widget type enables multi-dimensional analysis of your LLM application data, making it easier to uncover patterns and insights across different segments and metrics.
What are Pivot Tables?
Pivot tables provide a flexible way to summarize and analyze data by organizing it into rows, columns, and values. They excel at revealing relationships between multiple dimensions simultaneously, helping you answer complex analytical questions like:
- Cross-Model Performance: How do latency and cost metrics compare across different models and user segments?
- Feature Analysis: Which features drive the highest quality scores across different time periods?
- User Behavior Patterns: How do usage patterns vary by user type, geography, and application features?
- Quality vs. Cost Trade-offs: What’s the relationship between model costs and quality scores across different use cases?
Getting Started
Creating a pivot table is straightforward:
- Navigate to Dashboards → Widgets → New Widget
- Select your data source (traces, observations, or scores)
- Choose Pivot Table as your chart type
- Configure your pivot table:
- Metrics: Select which metrics you want to plot, e.g. different latency percentiles
- Dimensions: Select up to two dimensions that you want to group by
- Apply filters to narrow down your analysis
- Save and add to your dashboard
Learn More
For detailed instructions on creating and customizing pivot tables, visit our Custom Dashboards documentation.
Discover how pivot tables can transform your LLM application analytics and help you make more informed decisions.
Questions or Feedback?
We’d love to hear how you’re using pivot tables in your dashboards! Share your feedback and use cases in our GitHub Discussions.
Please open a GitHub issue if you encounter any problems or have suggestions for improvements.