Exploring a ChatGPT and Flowable Integration for Streamlined Processes

Nowadays everybody is talking about ChatGPT, but not everybody knows that the history goes back to the 50’s when the term Natural Language Processing (NLP) first came up. The current technology behind GPT: “Generative Pre-training Transformer architecture” was initially published in a paper from Google back in 2017. Since then, it still took five years until it got famous.

What is ChatGPT?

Behind the scenes there is a Large Language Model (also called LLM) which is basically a function to predict what is coming next based on the previous term. ChatGPT is supposedly using reinforcement learning from Human Feedback, using 100 trillion parameters (100.000.000.000.000) and 300 billion words. The large language model is then used with a certain random factor to improve the responses and not always take the mathematical best response. This makes the response human-like and often even better. OpenAI, the company behind ChatGPT, also offers a REST API which gives you access to GPT 3 and GPT 3.5. That allows you to integrate their service into your application and into Flowable on a native way. We considered different approaches and explained them in detail in our Webinar: Using ChatGPT to enhance case and process modeling. In this blog post we will summarize what we talked about.

Integration into the Flowable documentation

First, the search inside the Flowable Enterprise documentation changed. In addition, to the traditional search, there is now the possibility to use ChatGPT to do an aggregation of different pages. This uses the for the question most relevant sections of the documentation together with the OpenAI API to respond to the question as appropriately as possible.

Enhancing end-user experience

The real value of using ChatGPT with Flowable lies in its ability to enhance the end-user experience of Flowable. Given that cases and processes are frequently executed, each individual execution involves repetitive work that is performed multiple times. The time saved through process optimization accumulates across each instance during execution.

For the webinar, we considered different use cases and used ChatGPT for content generation and sentiment analysis as part of Flowable case and process models. But there are many other possible use cases, like classification of text, data extraction, structuring of text, and more. The scenario we used to try out an integration was a loan application. Eventually, once the loan is approved, we generated an email based on the user data. Rather than sending every customer the same email, the goal was to adapt the email to the customer. Usage of the provided data within the process is the key part of using a ChatGPT/OpenAI based integration. This is the point which allows you to make your application better than just simple templates you already have before, and you have less effort to maintain it compared to a template.

Adapting the writing to the age and nationality of a customer is something which ChatGPT can easily do and makes the customer appreciate the service even more. Watch the full webinar to see the demo.

Support during modeling

When it comes to modeling, the first and most obvious choice would be to generate an entire diagram using ChatGPT. However, we experienced some limitations about the BPMN standard in GPT which makes it hard to generate a complex diagram. While it’s certainly possible and easy to generate a simple two or three human task process, it’s not that easy to generate a model which fulfills a specific business need. Considering real-life use cases, typically the issue is not drawing the CMMN and BPMN diagram itself. Most of it is self-explaining and the training effort isn’t high. People usually face two barriers: First, figuring out what their business process is. This is often not trivial and here the diagram is often more helpful than a wall of text to talk about it. It would be a disadvantage to write a long text explaining the business process compared to what the diagram is. The saying “A picture is worth a thousand words” also applies here. Second, business analysts often struggle at the beginning with expressions or things which can’t be done in Flowable out-of-the-box. We can support during the technical details. However, this is something which ChatGPT can support with, since partially unstructured data can be converted into structured data. For details like scripts or expressions in a business process, those can be easily converted based on the meta information provided in the appropriate format expected in the editor. We are still experimenting with how powerful this is for complex scenarios. But for basic scenarios it works quite well and is reliable.

Summary

There are a lot of different ways ChatGPT or OpenAI can make your daily life easier. In this blog post we summarized what can be done. In case you are interested in further details how this could look like, we recommend watching the webinar. There are not only further explanations on how a business process model could look inside, but it also includes multiple live demos on how that behaves in practice.