How Greenlight Financial built an autonomous SOC solution on Amazon Bedrock AgentCore
Greenlight Financial’s Luna platform, built on Amazon Bedrock AgentCore, autonomously resolves over 90 % of up to 4,000 monthly security and risk cases by orchestrating a tiered multi‑agent architecture that incorporates episodic memory for continuous learning. The system layers agents—each specialized for distinct SOC tasks—while feeding new incident data back into a shared memory store, enabling the agents to adapt without manual re‑training. This approach demonstrates that a production‑ready, self‑learning SOC can scale to thousands of cases per month with minimal human oversight, though it requires careful design of agent responsibilities and memory integration to maintain performance.
⚡ Key Takeaways
- Luna achieves >90 % automated resolution on 4,000+ monthly security incidents.
- The architecture uses a tiered agent hierarchy on Bedrock AgentCore, with each tier handling specific SOC functions.
- Episodic memory is leveraged to continuously update agent knowledge, reducing the need for periodic model retraining.
- Deployment is performed via AWS CloudFormation stacks that provision AgentCore instances and memory stores.
- WhyItMatters: Engineers deploying autonomous SOC workflows can reduce incident triage load and accelerate response times by adopting a tiered, memory‑
Engineers deploying autonomous SOC workflows can reduce incident triage load and accelerate response times by adopting a tiered, memory‑
Want the full story? Read the original article.
Read on AWS ML Blog ↗