Unlock powerful AI functionality in your existing BRYTER applications
February 1, 2024
Steffen Huß
Integrating ChatGPT with BRYTER unlocks a wide range of use cases: You automatically can
classify texts,
extract text fragments,
reformulate,
do a first risk check,
and much more, all integrated into the decision logic of a BRYTER module. The approach described here uses a non-included add-on provided by BRYTER, but it is simple and quick to implement.
To begin, first activate the "AI Connector". This can be done easily from within your application in the "AI Connector" section. Click "Activate trial" to start immediately; additional charges will only apply after the trial period.
After activating the AI Connectors, begin building the module. In this simplified example, we'll create a module to classify company employees' legal requests, so that it can be redirected to the responsible person.
First, create a text input labeled "Matter Description," where users can detail their legal issue. Next, create an action and select "AI General Prompt with text input" from the options. This selection provides access to the latest model from OpenAI, which currently is gpt-4-32. The model's window size is set to 32k tokens, but to ensure prompt responses, the output is limited to 500 tokens.
Several other helpful options are available besides the ones mentioned. You can prompt on a file, for example, if the end user uploads a PDF. You can also use one of the many AI Connectors for specific purposes such as extracting a clause, summarizing a text, or checking a clause for compliance with internal rules. You can learn more about the different AI Connectors here. Although there is a "Categorize" AI Connector available, this post will focus on the "General Prompt" type.
Next, create your prompt. To improve readability, it helps defining the prompt in a Text Block action. Don't forget to reference the text to classify.
Then, reference your prompt Text Block in the AI Connector node. Optionally, you can add a system prompt.
Finally, you can use the answer given by ChatGPT in the module, e.g. by referencing it or by using it in conditions.
Be aware that a response from ChatGPT may take some time, which could result in your end user having to wait. Therefore, avoid adding several AI Connectors consecutively without an Input node in between, when possible.
