Eli Lilly's Model Gateway brings MLOps to pharmaceutical drug discovery at scale

Eli Lilly has deployed a custom machine learning operations platform designed specifically for the pharmaceutical industry's specialized needs in computational drug discovery. The Model Gateway addresses infrastructure gaps that general-purpose MLOps solutions cannot meet, managing version control for scientific models, orchestrating multiple AI systems for simultaneous parameter optimization, and enforcing governance aligned with how pharmaceutical organizations actually work.

Why pharma needs its own MLOps#

Generic ML platforms handle deployment and monitoring, but drug discovery pipelines impose distinct requirements. Scientists need to compose multiple models at inference time to optimize multiple chemical properties simultaneously—a task typical MLOps tools don't support. Physics-based computational models often lack serialized artifacts, complicating standard version management. Pharmaceutical companies maintain massive compound libraries that benefit from precomputation, and their governance structures revolve around scientific departments rather than generic user roles.

How the Model Gateway works#

The platform provides centralized control across heterogeneous clients—molecular design suites, large language models, and other downstream applications. Key capabilities include:

• Asynchronous execution for long-running simulations
• Consensus model orchestration to combine predictions from multiple models
• Automated retraining pipelines to keep models current
• Unified API service for diverse client applications
• Pharma-structured governance reflecting departmental and scientific organization

Scale in production#

Operating since deployment at Eli Lilly, the platform currently manages over 200 deployed models spanning small molecules, peptides, and antibody compounds. It serves more than five downstream applications across all phases of the drug development cycle: Design, Make, Test, and Analyze.