Prodigy

Prodigy is a commercial annotation tool for creating training data for Machine Learning (ML) workflows (machine learning / data annotation), developed and distributed by Explosion as part of its Natural Language Processing (NLP) tooling ecosystem.

• Scriptable annotation workflows for text, images, and other data types (data labeling).
• Integration with spaCy models for active learning, pre-annotation, and NLP-centric workflows (natural language processing).
• Web-based, lightweight interface for individual annotators and small teams (data annotation UI).
• Extensible configuration and plugin approach using Python scripts and recipes (developer tooling).
• Support for creating and managing labeled datasets for tasks such as text classification, named entity recognition, and other structured outputs (ML training data management).

More About Prodigy

Prodigy is a scriptable annotation tool (machine learning / data annotation) focused on enabling teams to create high-quality labeled datasets for ML applications, with an emphasis on NLP workloads. It is developed by Explosion, the company behind spaCy, and is distributed as a commercial, closed-source product that integrates closely with spaCy models and pipelines.

The core purpose of Prodigy is to streamline the process of generating training data by combining a Python-based configuration model with a browser-based annotation interface. Users define “recipes” (developer tooling) that describe how data is loaded, how models are applied for pre-annotation or active learning, and how examples are presented to annotators. This approach allows teams to use existing models to prioritize uncertain or informative examples, which can reduce the amount of manual labeling needed to train or refine models.

Prodigy supports multiple task types (machine learning / NLP), including text classification, named entity recognition, sequence labeling, and other structured annotation schemes exposed through its recipe system. The tool can also be used with images and other data modalities when defined via custom recipes. Annotations are stored in a backend database or files, and they can be exported into standard ML training formats or integrated directly into spaCy training workflows.

From an enterprise usage perspective, Prodigy runs as a local or server-hosted web application (application tooling) that annotators access through a browser. Organizations typically integrate Prodigy into Python-based data science environments, often alongside spaCy pipelines, to support iterative model development: labeling, training, evaluation, and refinement. Because recipes are written in Python, engineering teams can embed custom business logic, connect to internal data sources, or implement tailored model-assisted labeling strategies.

Technically, Prodigy operates within the broader Python ecosystem (developer tooling), aligning with common practices in ML engineering. It interoperates with spaCy for pre-annotation and training, and can be used with other frameworks via data export and custom scripts. The extensible recipe mechanism and component hooks provide a way to adapt the tool to internal Machine Learning Operations (MLOps) pipelines, versioned datasets, and experiment tracking systems maintained by enterprises.

Within a technical directory or catalog, Prodigy can be categorized as a data annotation platform for ML (ML tooling), with a particular focus on NLP and active learning workflows integrated into Python-based model development stacks. It is relevant to data science teams, ML engineers, and NLP practitioners who need controlled and scriptable tooling to build and maintain labeled datasets for supervised learning.