Perform Root Cause Analysis

This guide covers how to perform Root Cause Analysis (RCA) on the evaluation results of your LLM-based application using Deepchecks.

Deepchecks evaluates your application’s performance across multiple properties and overall, employing an end-to-end auto-annotation pipeline. You can learn more about it in the Configure Auto Annotation guide. The combination of these assessments offers strong signals for identifying the root causes of problems in your application, and Deepchecks provides a robust suite of features to support this process. You can view the full list of RCA features here, or use this guide to learn how to leverage them through step-by-step examples.

This guide has 3 steps:

1. Identifying Failures: Where you'll learn how to isolate the tasks and metrics where your application fails.
2. Analyzing Failures: Where you'll learn how to analyze those failures and think of mitigations for them.
3. Examples: Where you'll see end-to-end examples of the workflows described in steps 1 and 2.