Pega GenAI monitoring and analysis in PDC

The SQL AI Assistant combines rule-based analysis with Pega GenAI to identify common performance bottlenecks and provide guided performance insights and actionable optimization recommendations. The SQL AI Assistant is available on the AI Assistant tab in qualifying PEGA0005 cases.

The SQL AI Assistant evaluates each query by using the following sequence of checks:

• Problem origin: Determines whether a potential issue originates from Pega Infinity™ platform code, the application layer, or an invalid or suboptimal configuration.
• Presence of user-defined functions: Checks queries against external tables for user-defined functions that might degrade performance, especially when operating on large BLOB fields.
• Presence of performance-degrading operations: Checks for potentially inefficient SQL patterns, including wildcard searches, case-transformation functions, and sorting or grouping operations.
• Index structure: Reviews index metadata to determine whether existing indexes adequately support the query and might recommend index improvements or additions.

The SQL AI Assistant presents its findings in a structured format. The findings can include identified issues, root-cause explanations, index observations, optimization suggestions, and relevant constraints. Treat the suggestions as guidance for investigation and performance tuning.

The AI Assistant tab is available when all the following conditions apply:

• A PEGA0005 Case exists.
• The Case originates from a Slow Report or Slow RDB-List event.
• PDC associates one or more SQL queries with the Case.

PDC applies usage limits to Pega GenAI analysis. When the PDC reaches the usage limit, the Generate button remains unavailable until the limit resets. For more information, see SQL AI assistant for PEGA0005 slow SQL cases.

PDC provides operational visibility into Pega GenAI interactions and related service events. PDC collects execution data from Agents, Coaches, and Pega GenAI Connect Rules, enabling architects to monitor interactions, investigate issues, and assess the performance, stability, and operational health of GenAI-enabled applications.

This visibility helps teams identify performance, availability, and resource-usage issues.

PDC monitors alerts that relate to the availability, performance, and resource usage of application capabilities that use Pega GenAI. Application teams can configure thresholds for supported monitoring categories based on application requirements. When an interaction exceeds a threshold, PDC publishes contextual information, such as the Rule execution details, the model used, and elapsed time, to support monitoring and debugging.

PDC captures the following alerts generated from Pega GenAI capabilities:

• PEGA0175: Pega GenAI model nearing its deprecation date.  Alerts teams that a configured GenAI model is approaching retirement and should be replaced with a supported model. The system raises the alert 90 days before the model deprecation date.
• PEGA0176: Pega GenAI model no longer available. Alerts teams that a configured GenAI model is no longer available and may affect GenAI-enabled application functionality.
• PEGA0181: Run time of a Pega GenAI Rule exceeds the specified threshold.
• PEGA0182: Run time of a large language model (LLM) exceeds the specified threshold.
• PEGA0183: Excessively large request is sent to the LLM or multiple users send too many requests to the LLM within a minute.
• PEGA0184: A Pega GenAI Rule attempts to run a request that exceeds the token limit specified for the selected LLM.

These alerts help architects proactively identify model lifecycle issues, performance bottlenecks, model constraints, and integration issues that affect GenAI-enabled applications.

Architects can use monitoring data, alerts, and correlated events to identify issues that require further investigation in the GenAI Improvements landing page.

The Improvement Plan -> GenAI Improvements landing page presents the TOP 5 improvements across GenAI interaction, performance, and stability categories. The landing page helps teams focus on the high-impact GenAI issues and optimization opportunities instead of investigating individual events separately.

The GenAI Improvements landing page displays the most impactful GenAI cases with at least one event in the last 72 hours.

The landing page organizes GenAI cases into the following categories:

• GenAI Interactions: Highlights slow processing time for GenAI interactions and helps identify user-facing latency issues.
• GenAI Performance: Highlights performance bottlenecks within GenAI components, such as Agent processing, LLM execution, and related services.
• GenAI Stability: Highlights errors, exceptions, and runtime issues that affect the reliability and stability of GenAI-enabled capabilities.

For each category, PDC surfaces the most impactful cases together with trend information, including Web impact, Service impact, Background impact, Total events, Trend, First seen, and Last seen. This information helps architects prioritize remediation efforts based on business impact, frequency, severity, and recurrence patterns.

The GenAI Improvements landing page also supports application-level filtering, enabling architects to focus analysis on a specific application or review GenAI issues across the entire environment.

PDC aggregates related GenAI events into Cases and highlights the most significant issues in the GenAI Improvements view. Lead System Architects can use this view to:

• Monitor Pega GenAI performance trends
• Identify recurring GenAI failures
• Investigate root causes by using correlated events
• Prioritize improvements based on impact and frequency
• Track the results of remediation actions over time

For more information, see Improvement Plan landing page.

Check your knowledge with the following interaction: