To better manage vast regulatory data and streamline AI workflows across eight centers, the FDA developed a unified generative AI platform built on a centrally governed Databricks foundation, resulting in rapid staff adoption and significant efficiency gains in drug and device review processes.
- Centralized Databricks platform cuts data sharing latency from days to real-time
- Governance via Unity Catalog ensures secure access to sensitive regulatory data
- Scalable AI agent creation empowers diverse staff, raising platform use to 85%
Infrastructure signal
The FDA replaced fragmented AI infrastructures across its eight centers by consolidating dozens of disparate data sources into a single Databricks-powered cloud platform. This centralization transformed batch-oriented data workflows into real-time data streaming, enabling faster, more integrated insights. The adoption of Unity Catalog added a comprehensive security and governance framework, addressing strict compliance requirements around sensitive regulatory and trade secret information through granular access controls at the table level.
By unifying data storage and processing into a governed cloud foundation, the FDA reduced duplicated infrastructure costs and increased reliability through managed services. The platform now supports seamless data sharing and centralized observability, which simplifies monitoring and troubleshooting of AI workloads. This infrastructure strategy sets a foundation for extensible AI-powered apps spanning multiple regulatory domains with secure, auditable data provenance.
Developer impact
Leveraging the MCP server layer atop Unity Catalog, the FDA democratized AI agent creation, allowing medical reviewers, scientists, and even administrative staff to build and deploy custom, contextual AI assistants without heavy coding or data science expertise. This has dramatically expanded the internal developer ecosystem and accelerated iteration cycles for AI applications tailored to center-specific data and workflows.
The integration with Databricks MLflow for machine learning lifecycle management further enabled productive collaboration between AI engineers and domain experts to extract key insights from millions of pages of regulatory submissions. This tooling reduced the time for complex queries—like identifying drug starting materials—from several days to mere minutes, fundamentally shifting staff workflows from manual information gathering to rapid decision support.
What teams should watch
Teams focused on regulatory review, data science, and IT operations should monitor the ongoing expansion of the unified AI model framework from the drugs center (CDER) to other FDA centers, as it involves adapting tools to diverse data contexts and compliance landscapes. Observing how governance policies evolve to balance secure access and collaborative innovation will also be critical for scaling AI responsibly in regulated environments.
Additionally, development groups building or integrating AI platforms must track the user-driven proliferation of hundreds of custom AI agents weekly, which may introduce new deployment, monitoring, and versioning challenges. Emphasizing observability and lifecycle management practices tailored to non-technical users will be key to sustaining reliability and performance as usage scales.