How HPE Zerto built an agentic troubleshooting system with Amazon Bedrock
HPE Zerto has engineered an on‑premises, agentic troubleshooting system that leverages Amazon Bedrock. The solution deploys a multi‑agent architecture using Strands Agents to run Bedrock models inside the customer environment, enabling real‑time grounding of agents against live operational data. The post outlines the challenges of maintaining agent fidelity in a dynamic on‑prem context and the design decisions required to keep the system responsive and secure. The result is a tightly coupled, deployable stack that brings Bedrock’s generative capabilities into enterprise infrastructure without exposing data to the cloud.
⚡ Key Takeaways
- The system is built on Amazon Bedrock, providing Bedrock’s foundation models for agent reasoning.
- Deployment uses Strands Agents, a framework for running Bedrock agents on‑premises within customer environments.
- Grounding agents in live data requires custom integration layers to keep the agents’ knowledge base up‑to‑date.
- The architecture supports a multi‑agent setup where individual agents specialize in distinct troubleshooting tasks.
- WhyItMatters: Engineers deploying generative AI in regulated or latency‑sensitive environments can now embed Bedrock‑powered agents directly on‑prem, preserving data locality while still benefiting from cloud‑scale model capabilities.
- TechnicalLevel: Intermediate
- TargetAudience: ML Engineers
- PracticalSteps:
- Deploy Strands Agents on your on‑prem infrastructure to host Bedrock agents.
- Configure the grounding layer to feed live operational data into the agents’ context.
- Validate agent responses against a set of known troubleshooting scenarios to ensure fidelity.
- ToolsMentioned: Amazon Bedrock, Strands Agents
- Tags: AGENTS, DEPLOYMENT, AMAZON
🔧 Tools & Libraries
Engineers deploying generative AI in regulated or latency‑sensitive environments can now embed Bedrock‑powered agents directly on‑prem, preserving data locality while still benefiting from cloud‑scale model capabilities.
✅ Practical Steps
- Deploy Strands Agents on your on‑prem infrastructure to host Bedrock agents.
- Configure the grounding layer to feed live operational data into the agents’ context.
- Validate agent responses against a set of known troubleshooting scenarios to ensure fidelity.
Want the full story? Read the original article.
Read on AWS ML Blog ↗