Apps
Integration of 3rd party Apps
Last updated
Was this helpful?
Integration of 3rd party Apps
Last updated
Was this helpful?
In the Apps Step of Flow Builder, users can access third-party integrations represented as Apps. Currently, we have ChatGPT as the default App.
Introduction
Picky Assist's Flow Builder is a user-friendly tool designed to make your life easier. It helps you save valuable time by allowing you to effortlessly create personalized and dynamic chatbots for any industry, all without coding. However, when it comes to advanced capabilities, ChatGPT stands out. As a powerful generative AI, it offers superior performance. By integrating ChatGPT with Flow Builder, you can enhance your chatbots significantly. This integration allows you to direct user queries to ChatGPT, which, in turn, provides more natural and improved responses within the conversational flow itself. This means your chatbots can now offer even better interactions, making your user experience seamless and engaging.
Picky Assist's ChatGPT integration operates based on the Apps in Flow builder. You can add the ChatGPT App according to your requirements inside the Flow itself, providing better control over the conversational flow. This enables you to route customer questions to ChatGPT.
For each ChatGPT App inside the Flow, you can configure ChatGPT with System Data (instructions to ChatGPT) and Assistant Data (training data). For example, if you are building a Conversational Flow for a Dermatologist, you can set the behavior of the App as a Dermatologist by configuring the System Data. Additionally, you can provide training data about Doctor Consultation in Assistant Data. After configuring these settings, you can set up Message History. This feature allows you to pass message history to ChatGPT for more contextual information. For instance, if you have set it for the last 5 messages, the ChatGPT App works and delivers responses based on the patient's questions. After the 9th question from the patient, it will automatically send an appointment scheduling message for doctor consultancy. This can be achieved using the Formatter > Trigger Counter Action inside the Flow itself. Using this Conversational Flow, doctors and patients can easily save time and travel expenses, making the interaction more lively.
Create a Flow, Then Right Click on the Canvas, Then Select Apps (see the below picture)
Now You can See ChatGPT App action block inside the flow, Here you can Configure the Basic Settings of the ChatGPT App such as Default model, System(Set behavior),Assistant(Training Data),Enabling and Disabling Message History, Maximum Number of Tokens, Setting Temperature ,Saving ChatGPT response to variable etc
It is crucial to set a Default Model for the ChatGPT App because it defines how the chatbot behaves. For example, if you specify 'gpt3.5' as the Default Model, the chatbot will operate using the capabilities and characteristics associated with that specific version
Simply select the default model for your ChatGPT App (see the picture below)
Setting the system behavior in ChatGPT is like giving the chatbot a personality guide. It helps the chatbot know how to talk and respond, whether it should be serious and professional, or friendly and casual. This way, the chatbot can interact in a way that suits the situation or the user's needs.Click here to Read more
Assistant training data in the ChatGPT App involves teaching the chatbot by providing examples of conversations. Here, users can contribute specific questions and answers, enabling the chatbot to learn how to respond effectively. This process helps the chatbot engage in meaningful and accurate conversations with customers. Click here to Read more
Message History in ChatGPT is indeed like the chatbot's memory of the conversation. It allows the chatbot to recall past interactions, helping it understand the context of the current conversation. Users can pass upto 100 messages or even access all messages from the last 24 hours. This feature ensures the chatbot can provide better, more relevant responses, making the conversation smoother and more natural for the user. Click here to Read more
In ChatGPT, "Temperature" is like adjusting the chatbot's mood. A low temperature makes it more serious and focused, while a high temperature makes it more creative. Users can set the temperature based on the type of conversation they want to have with the chatbot. Click here to Read more
Here, the user is able to save the response from ChatGPT to a variable, and the user can give it a name for easy identification. This helps the user utilize the response in various actions like sending messages, emails, etc.
Users can utilize the response from chatgpt in various actions by simply mapping the saved response variable.(see the picture below)
For example, imagine you're a chess master building a conversational chatbot for your students. They ask questions, and you can pass those questions to ChatGPT. You then send ChatGPT's response back to them as your answer. You can easily do this by Mapping the saved response variable in action blocks like send message,Send Email action etc(as shown in the picture above).
It's essential to set up the next step so that users can know if their submitted request was successful or unsuccessful. By configuring the next step, the chatbot can indicate to users whether their request to ChatGPT was successful (passed) or unsuccessful (failed). This information helps users understand the outcome of their interactions with the ChatGPT app.
"Match Keywords from Response" is one of the most useful features in a chatbot. This feature enables the chatbot to perform actions based on the response from ChatGPT. Users can add keywords and steps associated with those keywords. If the response from the ChatGPT app matches any of the keywords, the configured step against that keyword will be executed. Here are some examples of how you can use the "Match Keywords from Response" feature:
In many scenarios, even if the ChatGPT app is well-trained for the business, end users can change the conversation's context by asking unwanted questions outside the original topic. To prevent these types of unwanted or unaimed conversations, we can use the "Match Keywords from Response" feature in the ChatGPT app. For example, if you have a Chatbot trained as a Cricket Analyst (a bot that answers cricket-related topics only) and the user asks about Hollywood movies and actresses, which are irrelevant, we can use the Match Keywords from Response feature. By adding words that are not related to the context, we can create a specific path for that response containing those words.(Please See the Picture below)
In another case, a chatbot is configured to assist with all customer inquiries in your organization. At times, a customer may feel the need to speak with a human agent. In this scenario, we can train the bot to recognize when a customer asks for human assistance and by adding keywords in the "Match Keywords from Response" feature. For example, we can add words like "Human," "Agent," etc., in the keywords, separating them with commas. From these keywords, we can route the chat to a human agent using the "Assign Chat" in action step.
In a case where the customer is not interested in continuing the conversation with the bot and requests to stop, here we can train the bot to recognize when a customer asks to end the conversation. And we can add keywords such as "Stop," "End," "Quit," etc., in the "Match Keywords from Response" feature, and then route these keywords to an "End" step to effectively end the conversation.
The user can enable and disable exact match in keywords.(see the picture below).For example, if the user sets a keyword like "End" to exit from the conversation and enables exact match (case sensitive), the system will look for the exact match of the word "End" in the response from ChatGPT.
If the user disables the exact match in the keyword (case insensitive), the system will look for whether the response contains the word or not. For example, if the user sets the word "end" as the keyword and if the response contains the word "send," the system will still execute further actions.
