Stacks/Internal engineering platform
Natera SWARM lab-operations agent
Natera built SWARM on Amazon Bedrock AgentCore to diagnose lab instrument tickets: a background agent drafts a report on every Jira ticket and an interactive runtime supports engineers, both drawing on a 33-skill library and a RAG knowledge base of vendor manuals. First response fell from 17.4 hours to about 8 minutes, with 96.2% of tickets covered automatically.
- Use case
- Diagnosing lab instrument support tickets for Natera's cfDNA testing labs: a structured diagnostic report on every Jira ticket before an engineer opens it, plus interactive troubleshooting
- Category
- Internal engineering platform
- Team size
- Natera lab engineering team led by Ariel Jirau, with AWS
- Orchestrator / framework
- Two AgentCore runtimes share one versioned skill library (33 skills: 24 instrument-specific, 9 workflow). swarm-agent selects skills by equipment type and symptoms, queries the knowledge base, and writes an HTML report (lead cause, ranked steps, confidence) to S3 linked on the Jira ticket; swarm-hive serves interactive follow-up with image upload. Jira events flow through Lambda to SQS; DynamoDB dedupes by ticket ID and stores per-run status, cost, and tokens. Deterministic work is Python; reasoning is LLM skills.
- Memory
- Amazon Bedrock Knowledge Bases (managed RAG) over a custom-parsed vendor documentation corpus, chunked with tiktoken cl100k_base, embedded with Titan.
- Evals
- Continuous loop: engineers record what they found, an evaluator grades each response Accurate, Partial, or Wrong by instrument family, fixes (usually knowledge-base updates) are re-run on the eval set. 524 tickets evaluated; High-confidence labels 74% accurate (154 cases). Adding missing manuals lifted one family from 71% to 75%.
- Observability
- Searchable CloudWatch logs per run, AWS X-Ray tracing from ticket intake to report with time and cost per tool call, failure alerts by email, and a Slack activity tracker. Every invocation is audit-logged.
- Cost (stated in writeup)
- Not stated
- Cost per output (stated)
- Not stated in the writeup
- WhatWeRun estimate
WhatWeRun estimate
Not enough info to estimate
Volume is stated (65+ tickets a day) but the LLM isn't named and tokens per ticket aren't given.
Reviewed Nov 13, 2026
- Observability tools
- Other tools
- JiraAWS LambdaAmazon SQSAmazon DynamoDBAmazon S3GitLab CI (OIDC keyless deploys)Slack
- Source
- https://aws.amazon.com/blogs/machine-learning/how-natera-designed-an-agentic-ai-system-for-lab-ops-with-amazon-bedrock-agentcore/
Ariel Jirau (Natera), Tamara Gagliardi, and Veronica Rule (AWS) · Published Nov 12, 2026 · Last verified Nov 13, 2026
Related
More like this
engineering.monday.commonday.com Infra Agents Platform
LLM agents that handle Infra tickets from R&D: reading and proposing changes to Terraform, Helm, GitOps, and CDKTF, reviewing infrastructure PRs, and resolving or escalating tickets
blog.duolingo.comDuolingo's production agent platform
An internal platform so any Duolingo team can define a production agent once and invoke it from Slack, internal sites, a CLI, or another Temporal workflow. In use for fixing CI failures, addressing code-review comments, and a Slack bot for release managers that investigates crashes and summarizes findings