Datadesigner, by Nvidia Nemo, is an mcp category orchestration framework built primarily with Python. It specializes in generating high-quality synthetic data, supporting Tool Use and MCP. Distributed as an open-source library, it integrates with the NeMo Platform as a service and connects to various LLM inference servers. This tool enables developers to define dataset schemas and scale synthetic data pipelines, providing a flexible solution for data creation.

What does Datadesigner connect to?

Datadesigner functions as an orchestration framework for synthetic data generation, categorized under Tool Use & MCP. It acts as a client to various Large Language Model inference servers, including NVIDIA, OpenAI, and vLLM endpoints, without hosting models itself. It integrates as a service with the NeMo Platform and can be wrapped by NeMo Curator for scaling. Additionally, it connects to external tools during generation via the Model Context Protocol, utilizing Files API, Secrets API, and Inference Gateway API from a NeMo Services cluster.

Core features and tools

Datadesigner exposes configuration-building tools within its "data_designer.config" package. Key components include ModelConfig, SamplerColumnConfig, and LLMTextColumnConfig, which are used to define dataset schemas. The DataDesignerConfigBuilder constructs these configurations. The DataDesigner.preview() method is available for generating new columns. Command-line interface commands such as "data-designer create" are also provided for execution. It enables LLM columns to call external tools via MCP for tasks like lookups or retrieval.

Who is it for?

Datadesigner is for developers and data scientists who need to generate high-quality synthetic data and orchestrate LLM-driven pipelines. It benefits those integrating with the NeMo Platform or scaling synthetic data efforts with NeMo Curator. Users requiring fine-grained control over dataset schemas and external tool interactions via MCP will find it valuable. This is not the right fit if you require an all-in-one solution that hosts LLM models internally.

How to get started

Access is granted through a standard OAuth authorization flow for the NeMo Platform service. For LLM providers, the server authenticates using an API key passed as an environment variable, such as NVIDIA_API_KEY, via HTTP.

Use cases

• Generating synthetic data for model training and testing.
• Orchestrating complex data pipelines involving multiple LLM inference servers.
• Defining and managing dataset schemas with LLM-generated columns.

Datadesigner by Nvidia Nemo offers a Python-based orchestration framework for synthetic data generation within the mcp category. It supports a diverse range of LLM inference servers and integrates with the NeMo Platform, allowing for scalable data pipeline development. Its configuration tools provide granular control over dataset schema definition and external tool interaction. Try generating new columns using the DataDesigner.preview() method to explore its capabilities.