nemo-data-designer-plugin

nemo-data-designer-plugin is an NVIDIA (NeMo Platform) skill for creating synthetic datasets and data-generation pipelines using the Data Designer library. It activates when a user wants to create a dataset, generate synthetic data, or build a generation pipeline, and it drives an end-to-end workflow that produces a runnable Python script defining a load_config_builder() function that returns a DataDesignerConfigBuilder. The problem it solves is turning a natural-language dataset description into a validated, previewable configuration without the user needing to memorize Data Designer's config schema.

The skill offers two modes: Autopilot (make reasonable decisions autonomously when the user signals they don't want questions) and Interactive (the default, driven by a separate workflow file). The documented workflow resolves the nemo CLI, learns available context via nemo data-designer agent context, infers diversity axes and schema, plans columns/samplers/processors/validators, writes the script with PEP 723 inline dependency metadata, then validates, previews with saved sample records (surfaced as clickable HTML), and optionally creates a full dataset of N records. Reference files cover person sampling, seed datasets, preview review, and NeMo Platform plugin additions such as sourcing model configs from IGW providers. The skill is explicit about safety and permission: it will not install the CLI or dependencies without the user's permission, and its troubleshooting guidance asks before retrying commands with a sandbox disabled. Its skill card notes no external credential is required and advises against putting secrets in prompts or logs.

Target users are developers and ML engineers who need synthetic training data, evaluation datasets, or reproducible data pipelines, especially those working within the NVIDIA NeMo Platform ecosystem. It requires Python 3.11+ and the nemo data-designer CLI.