Background
Relation Therapeutics adopted a standardised MLOps framework on Amazon SageMaker to streamline ML workflows, enhance model governance, and accelerate scientific discovery with scalable, secure, and reproducible infrastructure.
The Challenge
Relation Therapeutics needed to modernise and scale its machine learning workflows to accelerate drug discovery while maintaining governance and reproducibility.
- Fragmented ML workflows built on ad-hoc scripts and unmanaged infrastructure slowed research progress.
- Manual processes across data preprocessing, model training, and deployment led to inefficiencies and reproducibility challenges.
- Lack of standardised governance limited auditability, model traceability, and regulatory compliance.
- Collaboration across research teams was hindered by inconsistent tooling and access controls.
Our Solution
A standardised MLOps framework on SageMaker was proposed to automate and govern Relation Therapeutics’ ML workflows:
- End-to-end automation using SageMaker Pipelines for data prep, training, evaluation, and deployment
- Integrated data flow via Amazon S3, AWS Glue, and Feature Store for consistent, reusable features
- Governed ML lifecycle with Model Registry, Amazon CloudWatch, and AWS CloudTrail for versioning, monitoring, and audit
- Secure, scalable operations with IAM-based access control and real-time/batch model deployment
The Result
The implementation of standardised MLOps on SageMaker has enabled Relation Therapeutics to achieve:
- 3x faster model development cycles, reducing iteration time from months to weeks
- 70% improvement in model reproducibility and auditability through automated, versioned pipelines
- 50% reduction in manual effort via automated data preprocessing, training, and deployment
- Improved model performance tracking using integrated monitoring and evaluation tools
- Enhanced compliance and governance with a centralised model registry and audit logging
- Seamless collaboration across teams with secure, role-based access to ML workflows
