Principle:Truera Trulens Dashboard Visualization
| Knowledge Sources | |
|---|---|
| Domains | Observability, Visualization |
| Last Updated | 2026-02-14 08:00 GMT |
Overview
A visualization pattern that launches an interactive web dashboard for exploring application traces, feedback scores, and comparison leaderboards.
Description
Dashboard Visualization provides a Streamlit-based web interface for exploring TruLens evaluation data. The dashboard displays:
- App Leaderboard: Comparison of different app versions by aggregate feedback scores
- Record Explorer: Individual trace inspection with span-level detail
- Feedback Results: Score distributions and detailed reasoning for each evaluation
- Trace Viewer: Visual representation of OTEL span trees
The dashboard runs as a separate process, reading from the same database backend configured in TruSession. It supports multiple deployment modes including local, Streamlit-in-Snowflake (SiS), and Snowpark Container Services (SPCS).
Usage
Use this principle after recording application traces and running feedback evaluations. Launch the dashboard to visually inspect results, compare app versions, and identify quality issues. The dashboard is particularly useful for iterative development where you want to compare multiple experiment runs.
Theoretical Basis
Dashboard Visualization follows the Observer-Dashboard pattern common in monitoring systems:
- Data Layer: Reads from the persistent store (SQLite/Snowflake) configured in TruSession
- Processing Layer: Aggregates scores, computes statistics, formats traces
- Presentation Layer: Streamlit UI with React custom components for trace visualization
The dashboard is decoupled from the evaluation pipeline — it reads persisted data and can be launched independently of the recording process.