Troubleshooting AI Agent Reasoning Framework

• Context Gathering: The AI agent dynamically gathers contextual data when invoked. The agent collects context by querying incidents, alerts, and other metadata.
• RAG Integration: The Troubleshooting Agent is designed to troubleshoot dynamically and root cause a wide variety of network operational use cases like configuration errors, log analysis, link quality issues, route reachability issues, and interface errors. The AI Agent uses Retrieval Augmented Generation (RAG) to retrieve the product documentation, knowledge base, and product architecture information, thus enhancing the accuracy of the Agent.
• Dynamic Playbook Generation: The Troubleshooting agent dynamically builds a custom troubleshooting plan based on real-time context, for example, for an issue related to application slowness, the AI Agent will dynamically build a plan to check the underlay link circuit, application metrics, performance, and QoS policy configuration to then dynamically correlate multiple data points.
• Backend Tool Access: The Troubleshooting Agent then takes the action of calling the backend tools specific to the generated action plan. Multiple backend tools like site analyzer, QoS config analyzer, circuit and log analyzer, and application analyzer are invoked, and output from those tools is correlated at run time.
• Correlation across multiple data points: The agent correlates multiple data points in near real-time, including interface status, system logs, errors, and recent configuration changes.
• Accelerated Diagnostics: By concurrently running diagnostic steps across the entire SD-WAN fabric, the agent correlates static network policies with dynamic data plane traffic significantly faster than manual processes, thus lowering the Mean Time To Resolution (MTTR).
• Transparent Reasoning: Every conclusion reached by the Troubleshooting AI Agent is backed by verifiable backend data, giving admins full visibility to monitor the diagnostic process and validate the Root Cause Analysis (RCA).
• Fallback Guidance: If a root cause exists outside the SD-WAN environment (for example, a local switch port failure), the agent provides recommended next steps for adjacent infrastructure.

Key Performance Indicators

• System Health Metrics:
  • Uptime: Availability of the Agent and tools invoked by the Agent.
  • Latency: Latency per invocation and task completion (The AI Agent has read-only permissions within the system, so task completion will measure the time taken to arrive at the root cause of the issue).
  • Scalability: Monitor backend resource usage during peak query periods.
• Quality/Efficacy Metrics:
  • Accuracy: Root cause identification success rate, validated by network admin feedback.
  • Issue Complexity: Measured in terms of the number of backend tools invoked and the number of data points correlated by the Agent, with weighing mechanisms for each metric.
  • Efficacy per Product Area: Metrics per product sub-area, for example, routing/VPN/ configuration/QoS/Security policy, etc.

Quantifiable Improvements

Security and Governance

• Human Oversight: While the agent plans and identifies remediation, the final execution of changes remains subject to administrator approval.
• Explainable AI: Every diagnostic step is logged with a 'reasoning path', allowing NetOps teams to verify the agent's logic before proceeding.

Summary