Setting Up Guided Conversations using Generative AI
Guided Conversations powered by Generative AI can transform the way your organization handles inquiries, making interactions more intuitive, efficient, and helpful. Below is a simple walkthrough on how to set up a basic Guided Conversation using Generative AI within your Capacity knowledge base.
Getting Started
- Start by Creating a Dialogue: In your Capacity knowledge base, initiate the setup by creating a new dialogue. You might name this dialogue, for example, "Support Document Bot."
- Add an Exchange: Click on the blue "Add Exchange" button. Here, you will create an inquiry to trigger the dialogue. Label this inquiry with a phrase such as "I need help with support."
- Create a New Conversation:After setting up your inquiry, proceed to create a new conversation linked to it.
- Replace the Multiple Paths Card with a "Collect Info" Card: The first card to include should be one that collects information from the user. Use a prompt such as "How can I help you?" and create a variable to store the response, which might be labeled as "question."
Setting Up Vector Search
1. Utilize the "Search Vector" Card:
- This card is your gateway to searching through your knowledge base using the query collected from the user.
- By default, it is set to search all documents, but you can limit this to specific folders for a more targeted search. For instance, set it to search only within your "Customer Support" folder.
The output of Search Vector card will include relevant document excerpts, which are passed to the AI Assistant Card (Large Language Model). These excerpts are the segments of the document most pertinent to the query. This method of fetching document snippets to feed the results into a large language model is known as "retrieval-augmented generation."
2. Narrow Down Search Results:
- Specify how many results to return. For the internal model, a recommendation is to keep it to three or four results, but for the external model, you could increase this to five or six results.
Integrating the AI Assistant
1. Select the AI Assistant:
- This is where you choose between Capacity’s internal AI or an external model—like GPT-4—for processing the search results.
- The Capacity AI Model has a limit of 2,048 input tokens so it could be best to limit vector output 2-3 results to ensure the model doesn't send an error.
- The option available will depend on your organization’s settings and whether you’ve elected to include external model functionalities.
2. Customize the AI Response:
- A default prompt is provided for handling inquiries effectively, but you have the freedom to customize this further.
- Whether you require responses in a different language, want bullet points, or seek a specific tone, you can adjust the AI’s replies accordingly.
Ensure "{{nlp.inquiry}}" variable is adjusted to include the question variable collected from the chatbot user.
3. Finalize with a Simple Message Card:
- Insert the output from the AI Assistant into a Simple Message card to communicate the AI-generated response back to the user.
3. Testing Your Setup
- Once you’ve configured the guided conversation, it's essential to conduct tests to ensure everything operates smoothly. Engage with the dialogue as a user would and assess how effectively the AI navigates and fetches the relevant information based on your queries.
Moving Forward
This guide provides a primer on setting up a basic guided conversation using Generative AI in your Capacity knowledge base. For more advanced workflows or specific needs, consider consulting with your client success manager. They can offer tailored advice and support to leverage Generative AI fully within your Capacity instance, enhancing your capabilities and streamlining user interactions.
Thank you for following this guide, and welcome to a new level of efficiency and intelligence in handling support inquiries with Capacity.