How Johnson & Johnson built a scalable NLP Suite on Domino

The R&D Quality team at Johnson & Johnson supports analytics across regulatory and clinical documentation, where much of the data is unstructured text. Extracting insights from that text had become increasingly difficult, with more than 10 disconnected NLP tools in use, each built in isolation and hard to scale.

Consolidating NLP workflows into one extensible platform

To modernize their approach and streamline how teams worked with unstructured data, the team created an NLP Suite using Domino as the foundation. Each NLP method, including text classification, causal inference, and entity harmonization, was developed as a standalone component with its own code repository and Domino project. The suite calls each NLP method as needed, then processes and displays the results through a shared front end.

The team designed the platform to support parallel development, collaboration, and governance across a highly regulated environment. Source control is managed through Bitbucket and GitHub, while Jira and Confluence support agile planning and coordination. Teams work across development, QA, and production environments with the flexibility to merge contributions from both internal and external collaborators. Components are reusable across departments, and everything runs within a compliant, version-controlled framework.

Key features of the NLP Suite architecture

• Shared pre-processing and visualization across NLP methods
• Modular components for plug-and-play extensibility
• Domino orchestration with shared result interface
• Integration with Bitbucket, GitHub, Jira, and Confluence
• Access management via Domino Organizations

From legacy app sprawl to unified NLP

The suite consolidated not only technical workflows but also user communities. More than 200 legacy users have transitioned to the NLP Suite, where shared access to tools and methods now supports cross-team collaboration and insight generation. Teams that were previously siloed now operate from a single front end and version-controlled codebase.

This shift has also reduced support overhead. The team estimates saving more than 800 hours annually, while cutting manual effort for text analysis by up to 90 percent.

Results:

800+ hours saved consolidating app support

90% reduction in manual text analysis

200+ users in unified, compliant suite

Scaling the suite across departments

The team is actively expanding the NLP Suite by integrating additional methods and connecting it to department-specific data sources. Their goal is to create customized versions of the suite tailored to different business needs, all while maintaining the compliance and auditability required to move from proof of concept to enterprise production.

We’ve started thinking more like software engineers, and Domino gives us the tools to do it. With modular design, version control, and real testing workflows, we can deliver results at scale, even in a regulated setting.

Deepak Bandyopadhyay

Director of Data Science, R&D Quality