Delve Bio accelerates infectious disease diagnostics with Union.ai

Delve Bio spun out of UCSF with a mission to modernize infectious disease diagnostics using metagenomic next-generation sequencing. Their test analyzes all nucleic acids in a patient sample—detecting bacteria, fungi, parasites, and viruses simultaneously from a single cerebrospinal fluid specimen.

The diagnostic pipeline is extremely sensitive. With picogram-level nucleic acid inputs, Delve must distinguish true biological signal from assay or environmental noise—requiring deeply traceable, reproducible, and auditable computational workflows.

Initially, Delve packaged and executed inference pipelines in a single container. As pipelines multiplied, this approach broke down:

• Inflexible resource allocation
  
• Dependency conflicts across teams
  
• Slow cold starts and rising compute costs
  
• Difficult iteration and environment management
  
• Limited reproducibility for clinical and regulatory needs
  

As the sole bioinformatics engineer at the time, Brian O’Donovan sought an orchestration platform that could support clinical-grade reproducibility, enable rapid scientific iteration, and reduce the operational burden of managing infrastructure.

Having used Flyte before, Brian championed adopting a workflow orchestrator to scale Delve’s diagnostic pipeline development. After evaluating AWS Batch, GCP options, and self-managed Flyte, Delve selected Union.ai, citing reliability, support, and seamless integration with their AWS infrastructure.

Union relieved Delve’s engineering team from dealing with Kubernetes complexity—especially important for a small team balancing clinical systems, billing systems, and R&D compute.

Key Union capabilities delivering value:

Strong Typing & Reproducibility

Union’s type engine reduces data mismatches and runtime errors—critical for clinical traceability.

“It’s important for us to have strong guarantees about inputs and outputs. It makes our predictions more reliable.” — Brian O’Donovan

Composable, Auditable Workflows

Delve uses Flyte/Union to run parameter sweeps, in silico experiments, and alignment-stringency studies with full caching and auditability—ideal for regulatory documentation.

Efficient Resource Management

Breaking monolithic containers into modular tasks reduced cold start times, simplified iteration, and improved compute efficiency.

Ease of Onboarding & Fast Iteration

Union enables Delve to go from notebook exploration → production workflows in hours. New researchers onboard quickly and begin registering their own workflows within weeks.

“The learning curve’s not steep, but the yield curve is incredible.” — Brian O’Donovan

With Union, Delve transformed both developer productivity and scientific velocity. Workflows that previously relied on large, monolithic containers became modular, typed, and reproducible. Scientists can now explore ideas in notebooks and turn them into production-grade workflows the same day.

Union’s managed infrastructure eliminated the need for Delve to maintain Kubernetes clusters, freeing engineering resources to focus on clinical systems and core diagnostic logic.

By providing strong typing, transparent caching, and clean audit trails, Union supports Delve’s clinical and regulatory needs while enabling rapid R&D iteration. As Delve expands access to its metagenomic infectious disease test, Union.ai continues to serve as a foundational piece of its AI development infrastructure.