This article shares a Proof of Concept (PoC) and some practical experience with Pega GenAI capabilities. Initially, I planned to share only a summarization use case, but I decided to enhance the solution and present a more flexible framework that can be used for a variety of GenAI scenarios.
The PoC demonstrates how a Pega case type can invoke a Large Language Model (LLM) using a Connect GenAI rule. We applied a similar approach in one of our projects to summarize information from existing and previously completed cases.
Important Disclaimer:
This solution is not production-ready. It is intended as a starting point for experimentation, learning, and building your own GenAI use cases.
Solution Overview
The PoC consists of a dedicated case type responsible for orchestrating GenAI interactions and storing execution details.
Key Advantages
Generated with Pega Blueprint
The initial application was generated using Pega Blueprint, allowing rapid development and experimentation.
-
The Blueprint package can be accessed here: Please replace the pdf extention to blueprint.
Gen AI Data Operations - 20260707T093620.129 GMT.PDF (27.7 KB)
-
You can extend the generated application to support your own use cases and business requirements.
Independent Architecture
The solution is designed to be independent of a specific source system.
- It can be triggered from both Pega and non-Pega applications.
- It maintains its own data model for storing request and response information.
- The architecture can be adapted to integrate with multiple source systems.
Auditable and Extensible
The solution can be configured to capture and persist a variety of execution details, including:
- User prompts
- System prompts
- Model information
- Generated responses
- LLM usage statistics
- Token consumption
- Execution timestamps
- Additional audit information
This provides a strong foundation for governance, monitoring, and future enhancements.
Case Type Overview
Document Operations stage does not related for this artice. Ignore it.
Prompt Management
The PoC currently stores prompts in a dedicated Pega data object. However, the architecture is flexible and supports alternative approaches:
- External data sources
- Configuration files
- User-provided prompts
- System-generated prompts
- Dynamic prompt composition during case processing
This enables teams to adapt prompt management to their specific governance and operational requirements.
Prompt Configuration
Data Object
Connect Gen AI Rule Configuration :
System Prompt :
User Prompt
Data Model
The initial data model was generated using Pega Blueprint and then extended to support the requirements of this PoC.
The model stores:
- Request information
- Prompt configuration
- LLM execution details
- Response data
This provides a reusable foundation for building GenAI-powered applications within Pega.
LLM Statistics and Monitoring
The solution captures execution information returned by the Connect GenAI rule.
This implementation was inspired by an approach described in another article, which allows organizations to gather valuable operational insights such as:
- Model used
- Token consumption
- Request metrics
- Response metrics
- Execution details
Additional details can be found in the related post: GenAI Activity Logging
Execution :
Create a case with context :
Result :
Execution Details :
Same context with summary prompt :
JSON Output prompt :
