How Natera designed an agentic AI system for lab ops with Amazon Bedrock AgentCore
Natera built SWARM, an agentic AI system on Amazon Bedrock AgentCore that captures institutional lab-operations knowledge and deploys it at scale. This post walks through the full architecture, from knowledge base construction to deployment and evaluation, delivering about 130x faster first response and 96.2% automated coverage across support tickets.
At Natera, a leading provider of cell-free DNA testing, the reliability of laboratory equipment is paramount as any downtime disrupts sample processing. To tackle this challenge, the lab engineering team, headed by Ariel Jirau, developed SWARM, an agentic AI system for lab operations, using Amazon Bedrock AgentCore. This article outlines the architecture of SWARM, from knowledge base construction to deployment and evaluation, enabling similar patterns to be applied to other operational workflows.
One of the primary challenges faced by Natera was preserving institutional knowledge. With over 65 support tickets daily across three Jira queues and an average first response time of 17.4 hours, equipment downtime resulted in delayed lab samples and patient reporting. When senior technicians departed, their vast knowledge was lost, making it impossible to maintain a standardized playbook for troubleshooting.
To address this, Natera opted for Amazon Bedrock AgentCore, a managed and extensible service by AWS, to build SWARM. This decision was driven by three key requirements: processing high-volume support tickets autonomously, seamless integration with existing tools like Jira and Confluence, and robust security and observability features without the need to build custom infrastructure.
SWARM employs two agents: the swarm-agent, which autonomously generates structured diagnostic reports before a technician opens a ticket, and swarm-hive, a real-time interaction platform for engineers to upload images, explore alternative troubleshooting paths, and collaborate. The architecture of SWARM consists of four layers:
1. The event layer routes Jira tickets through EventBridge and SQS.
2. The agent layer hosts the agents and delivers diagnostic reports back to engineers.
3. The context layer draws on vendor manuals, historical tickets, and operational data.
4. The observability and delivery layer captures traces, logs, and metrics.
Security was a top priority for Natera’s architecture. They chose the swarms-agent and swarms-hive in a secure, managed environment with built-in identity management and monitoring, eliminating the need to build custom infrastructure. Keyless deployment was selected because SWARM operates in a clinical setting where credential exposure is unacceptable.
OpenID Connect (OIDC) and AWS Security Token Service (AWS STS) generate temporary credentials per deployment, ensuring no long-lived keys are stored. IAM enforces least privilege at the component level, while Amazon VPC with private subnets and VPC endpoints keep traffic off the public internet. AWS Secrets Manager automatically rotates the few credentials that exist on a 30-day cycle.
Written by urgent.news from AWS Machine Learning's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
