General ChatBot Pega Infinity 24.2.4

I would like to create a chatbot that will be available on the User Portal and, eventually, also within cases. The chatbot should be connected to our third-party external NLP services.

My initial thought was to use the Intelligent Virtual Assistant (IVA) functionality in Pega as the foundation. However, I cannot find documentation on how to integrate IVA with an external NLP engine.

I am also not sure if this is the right way to do it.

Could you please advise if IVA is the right approach for this requirement?

If not, what would you recommend as the best practice for integrating a chatbot with an external NLP service?

Thank you!

Can you share more detail about your scenario? What outcomes are you looking for? Why the requirement for your own NLP service?

Pega has evolved a bit since the IVA days. I’d say it has gone in two directions.

In a customer service context, Pega has moved to Pega Digital Messaging. It gives you messaging across a whole bunch of channels: including WhatsApp, web chat, mobile chat, Facebook Messenger, SMS. It provides a built-in chatbot/virtual agent capability. I have not personally heard conversations around connecting to 3rd party NLPs, but I am no expert here.

The other direction is around agentic AI. LLMs are so good at handing human language. I think they’ve really taken over, and NLP itself is perhaps a little too low level to deal with these days. Why work on building an engine when you get a full car?

In 24.2, you can create a “Coach” that appears on the user portal. You can chat with it, it gets you answers, you can plug it into our case as well. I’m sure there are better videos that can help, but I posted one agentic AI video where the user is chatting with the Coach. Fast forward to about 9 :50 in the video.

https://players.brightcove.net/1519050010001/default_default/index.html?videoId=6391532939112

With Agentic AI, you do have the option of pointing to your own models, though it’s just a lot more turnkey to use Pega’s supported models.

Hope this helps a little!