Replay
Overview
The Pega GenAI™ capabilities introduced in Infinity ’23 can help you build applications even faster by creating workflows, mapping integrations to back-end systems, creating test data, localizing your applications, and more.
In this webinar, Dennis Grady and Pankaj Rawal provide a technological overview of Generative AI, then explain how Pega's GenAI™ capabilities do the heavy lifting for you, allowing you to focus on innovation and oversight of your Pega applications.
Downloadable slides from this webinar
Questions & Answers
Below are the different groups of Q&A from the webinar; use the links to jump to the section of your interest.
- Pega GenAI - Models & Availability
- Pega GenAI - How it works
- Pega GenAI & Decisioning
- End User Use Cases
- Developer Use Cases
- Resources
- Suggestions & Feedback
Pega GenAI - Models & Availability
| Question | Answer |
|---|---|
| What are the source(s) of data on which the Pega GenAI model is trained? | We currently use OpenAI’s GPT 3.5 Turbo hosted on Microsoft Azure which is trained on the open internet. We will talk about some future plans later in the presentation. |
| Is there a plan for Pega GenAI available for customers who are outside Pega Cloud in future? | This is something that we have been discussing internally, no concrete plans as of now. For now, the GenAI gateway service is currently available exclusively as a cloud-based offering. It is only supported by GCS on Pega Cloud. |
| Is there a possibility to plug other LLM models? | As of now, we don't support plugging in other LLM models. We do have plans of working on custom LLM from Pega but that's still under planning. |
Pega GenAI - How it works
| Question | Answer |
|---|---|
| How does Pega GenAI determine the case life cycle and data model based on the Application? | It is based on the application and all the contextual data that's needed. For data model, it needs the Application and Case type context. For suggesting Case types, we are sending Application context. |
| While creating Case Type and stages of the application, does it traverse or walk through most of the Pega OOTB tables? | It’s only used at design-time to recommend stages and steps and would not have any production impact by using GenAI to create the starting template. The custom use case that Pankaj showed includes a runtime use case. We recommend that these are additive to experiences and expect occasional failures with human in the loop to review any results. |
| How does Insights handle a relatively huge amount of data? Does it have similar performance impacts as report definitions? | Insight use case is implemented independent of data it has. Basically, Pega GenAI comes back with a suggested query rather than the data itself. |
| So the query execution is handled similar to report / SQL execution as we have in OOTB Pega? | Insight generation in that example shown was querying data in elastic search but it can fall back to SQL depending on the join criteria required that elastic cannot handle. The Insights front end gathers its data via the /data API. That is, there is a Data Page associated with the data object / case type that you are exploring, which ultimately does the data retrieval. |
| As of now, Pega GenAI creates all the steps in stages as assignments. Can it create any other parts of the case type, like utility or sub flow based on the use case? | It only creates assignments in Infinity 23. In Infinity 24.1, we’re extending to include notifications and approvals and will continue looking to extend it to be more capable. |
Pega GenAI & Decisioning
| Question | Answer |
|---|---|
| Will Pega GenAI replace Pega Decisioning? |
It compliments Pega Decisioning as opposed to replacing it. Think of Pega Decisioning as a Left Brain and Generative AI as a creative right brain. Check out this article - https://www.linkedin.com/pulse/navigating-ai-landscape-alan-trefler/ |
| Is there any design rationale for using GenAI for test data generation instead of using Pega's Monte Carlo dataset generation? | There are a set of users where the monte carlo data set generation is more advanced. If you’re already using Monte Carlo generation successfully I’d recommend contunuing to do so but the Gen AI generation provides a quick and easy option for small data sets. |
End User Use Cases
| Question | Answer |
|---|---|
| Can we delegate Explore Data to BAs or Manager roles? | Insights are available as a standard landing page inside a Constellation portal and its available on App Studio. Therefore, it is available for BAs directly and you can make it available to managers or other persona in your app. The ‘Chat with Your Data’ capability simply creates an Insight for you. As a user, you can further change it where needed or save it and perhaps add it to a dashboard. Note that saving Insights requires a production ruleset in production. |
| Does Chat with Your Data work on existing data in the application, for example, you migrate an existing application to Infinity'23 with Pega GenAI? Or does it only work iwth sample data that was generated from scratch by Pega GenAI? | It works with any data. At Pega, we use it internally on our own Sales Automation and Agile Studio deployments with great results. |
| Does Pega GenAI have the capability to identify scenarios like credit card duplicate charges, fraud detection, and help with the data related to CDH campaigning? | It has absolutely. But we have to be careful in what data we expose to LLM servers vs using RAG technique. In fact, it is extremely effective in providing suggestions for all these use cases. |
| Business people execute reports and update/add columns as per their needs. As per Pega GenAI, can we also introduce data query in the portals? | It is indeed meant for end user portals. We agree: Introducing a conversational way to explore data is really powerful for business users. |
Developer Use Cases
| Question | Answer |
|---|---|
| Does the AI translation work on the UIKit or Cosmos app rulesets? | It picks up everything that the localization wizard includes when you run it. |
| While creating a case type, how is Pega GenAI different from using a template? | Think of Pega GenAI suggestions as a more intelligent and contextual template generated specifically for that use case. Templates in general are static with fixed set of stages or steps which may or may not be relevant to the exact business case. |
| Can Pega GenAI generate C# or Java code? It could also help generate activities to do more advanced Pega stuff. | Theoretically, that is possible. The amount of use cases is huge. You can expect more developer support in future releases. |
| Can we define which fields need to be masked in the Connect Generative AI rule? | Yes you can. See https://docs.pega.com/bundle/platform/page/platform/data-integration/masking-pii-for-gen-ai.html |
| Will Pega GenAI consider Pega Naming Conventions while for the Creation of Stages, Processes & Steps? | Yes. We encourage to try it out with various use cases and do provide feedback on same. |
| Is the Gen AI Rule type and related feature available for Constellation? | Yes, the rule type working in constellation apps and just interacts with application data. The demo showed during this session was on Constellation. |
| Can Pega GenAI help generate/improve the content of Chatbot or Digital Messaging? | The Pega Customer Service team has been working on a number of enhancements that leverage Generative AI including chat transcripts and agent training use cases. |
| Will organizations allow sending data to GenAI to get the responses? Are there concerns about data breaches? | Ideally, we should not even be sending any sensitive data to external LLMs. We should make sure to mask any sensitive data before sending it. As mentioned during the session, there are plans of locally hosted and custom trained Pega LLM in future. |
| The sample form data in my trial site gave me a lot of John Doe's and I noticed some of your records were duplicated. Is there a way to create more diverse sample data? | There is a non- Pega GenAI approach using Monte Carlo data set generation that could meet your needs. |
| What happens if integration response has an array or complex structure? | It can do a reasonable job if the local data object has similar structures defined. It should definitely be reviewed though as sometimes internal data models don’t map cleanly to rest responses. |
| What security aspects to keep in mind while enabling GenAI for end users? | Sensitive/personal (PII) data is most crucial thing to take care while talking to external LLMs. Pega highly recommends following a human-in-the-loop approach where generative AI can assist end users complete their work more efficiently without replacing them. |
Resources
| Question | Answer |
|---|---|
| How do I enable Pega GenAI in my environment? | See https://docs.pega.com/bundle/pega-cloud/page/pega-cloud/pc/learning-about/genai-in-pega-cloud.html |
| How do I try Pega GenAI for myself? | Request a Pega Community Edition. These instances are Pega 23 with Pega GenAI enabled by default. Request your instance here - https://community.pega.com/get-started/community-edition |
| Can you recommend any course or material online for Prompt Engineering? |
https://www.promptingguide.ai/techniques/consistency This one is pretty cool. You can find many YouTube videos as well. The best practices that we shared during the session is based on our own learning and experiences. We want to share everything we learn with you, and hope you do the same to enrich the collective knowledge of the entire Pega Community. |
| I'm not able to find a lot of info on Vector DB. Is that a generic concept or is it Pega based? | It's a generic concept. You can find many articles on it. Check out https://medium.com/@raresalm/title-the-significance-of-vector-databases-in-generative-ai-eece823428e1 |
| Do you know when Pega Academy Challenges will include the Pega GenAI features? The sample data feature would be great there! | We are working closely with Academy on exposing all new features. We are not sure of exact timelines at this point in time. |
Suggestions & Feedback
| Question | Answer |
|---|---|
| Does Pega have a plan of enabling users to make use of json schemas? REST interface dm creation with request and response data is so tedious when you want to generate multiple methods with different attribute structures. | Great feedback. We do support OData today which does a bit of what you’re describing. I will need to follow up with our integration product team on where JSON Schemas fall on our roadmap. |
| Does Pega GenAI have the capability to generate sample content relevant to the end user when processing a case, for example, a PDF or image? | Not as of now. Feedback taken. Will definitely incorporate this in our planning and prioritization discussions. Great suggestion :) |
| Does Pega plan to create data for REST services using Pega GenAI in future? | Not as of now. Feedback taken. Will definitely incorporate this in our planning and prioritization discussions. Great suggestion :) |
| Can Pega GenAI be used to generate simulations for the connectors? | Not as of now. Feedback taken. Will definitely incorporate this in our planning and prioritization discussions. Great suggestion :) |
| Can it be used to fill the rule history and also the test case creation in later future based on rule name? | We are having some great discussions around history and automated test cases, not just for design time but for runtime as well. |
| Does Pega has any plans to provide connector for GraphQL? | Not as of now. Feedback taken. Will definitely incorporate this in our planning and prioritization discussions. Great suggestion :) |
| Suggestion: Use Pega GenAI to convert the section rule into views for all old apps. | This is something that's already in roadmap and prioritized. Thanks for sending this over. Will pass on this message to the teams :) |
| Pega GenAI is currently not generating persona and data. Is that coming as an enhancement in the future? | That's correct. It's on roadmap. Right now Persona suggestions are limited only to Persona landing page. Great feedback around Data objects generation. |