generate-synthetic-data

Creates realistic synthetic test inputs that target likely failures in an LLM pipeline.

When to use it

Use when bootstrapping an evaluation dataset, real user data is sparse, or specific failure hypotheses need testing; avoid synthetic generation when 100+ representative real traces already exist or the task is collecting production logs.

Give it an application description; it develops realistic scenarios with you, generates synthetic queries, and captures traces from running them through the pipeline.