The Knowledge base builder tool allows you to build a conversation easily via an intuitive user interface and then using use the resulting JSON in the knowledge base Skill. You can access the KnowledgeBase Conversation Builder from the left navigation menu of the Knowledge Base Skill or access it directly from https://knowledgebase-ui.prod.soulmachines.cloud/ .
From the landing page of the Knowledge Base Builder, you can either create your conversation from scratch or update/edit your existing Knowledge Base json file.
Creating new knowledge Base
Knowledge Base contains your library of Q&As: questions the user might ask around a given topic and how your Digital Person would answer. Start creating your base from scratch. Easily build your Q&A base with this tool, then train the NLU and test how it would play out. When you're done, collect export the Knowledge Base as a JSON file and take it back to DDNA Studio.
Note: The data doesn’t save, so make sure to collect the knowledge base before refreshing or changing pagesAccessing Knowledge Base Builder
You can access the KnowledgeBase Conversation Builder from the left navigation menu of the Knowledge Base Skill or access it directly from https://knowledgebase-ui.prod.soulmachines.cloud/.
Use your Soul Machines credentials to log into Knowledge Base Builder. The landing page of the Knowledge Base Builder lists all the Knowledge bases created by users of your organisation. You can either create your conversation from scratch or edit/duplicate existing Knowledge Base.
Creating new knowledge base
To create your Knowledge Base from scratch, press the + button and enter a name for the Knowledge base. Under each topic you can have multiple related questions and answers.
Field | Description |
|---|---|
Topic Name | Used as a unique identifier for each JSON object. (i.e. WhatIsSoulMachines) |
User Question/s | List of strings that contain examples of different utterances from the end user (i.e. "what is soul machines", "tell me about soul machines", "who is soul machines", "who are soul machines", "what are soul machines", "what is soul machine", "who is soul machine", "who are soul machine" ) |
DP Answer | List of responses the Digital Person should use to answer that question. If more than one value is found for a given topic, the skill will randomly pick one of the responses to reply with. (i.e. |
Each answer can be associated with visual elements using content cards and can also include gesture markups to enhance the Digital Person’s behavior. |
Adding Gestures Markup
Digital Person’s behavior can be adjusted using hashtag gesture markup inserted directly into the DP answers.
We offer following types of gesture hashtags:
Hand and body gesture markup - These can trigger unique movements of the hand and body that aren't automatically available in real-time gesturing.
Head and face gesture markup - These can trigger specific facial gestures and emphasise emotions in speech.
For detailed list of supported Gesture Markup, see Gesture Markup.
Adding Content Cards
In order to enhance the responses with media or links you can add content cards to the DP Answers. Available content cards are Image(s), video, link, list, and markdown. To add content cards:
Create a DP answer
Click the answer to Edit
In the dropdown select a content card you would like to use.
Training and Testing your Conversation
Once you have either created a new topic, or modified an existing topic, it will enable the “Train Conversation” button. It is necessary to train the conversation before you can test it. Training the conversation takes between 15 seconds and a minute, depending on the size of your knowledge base. During this time you’ll see the “Training …” button. At any time, you can switch from conversation mode to view the generated JSON using the toggle at the top right-hand corner.
Once the conversation is trained, you can start testing it by sending in chat messages that would simulate what a user might send. Any cards that are configured in the response will be shown, and the @ShowCards(xxx) command will be stripped from the response to replicate what a user would actually see.
Note: If you make a change to the knowledge base, the text input will go away and you’ll be required to re-train the conversation before you can continue testing.
Each of the content cards have different options as below:
Content card type | Field | Description |
|---|---|---|
Image | Image URL | When using a URL, the url must be accessible by the UI. For example, if the UI is accessed via https, then image URLs must also be accessed using https. Use the prefix The URL must have a prefix. |
Image Alt Text | An alternative text describing the image. | |
Video | Youtube Video Id | The YouTube videoId for the video you wish to embed. |
Automatically play video full screen | When autoplay is turned on the video will pop up to full screen and begin playing as soon as showcards is triggered for the card. | |
Close the card when video ends | When autoclose is turned on for a video card, the video player closes at the end of the video and enables you to immediately continue the conversation. | |
Option List | Label | Label is used to display text within the button / option in the UI. |
Value | Value may also be used to match a possible spoken response by the user. The list option could then be clicked as a form of selecting it, in addition to speaking the list option. | |
Link | Link Type |
|
Title | Text that is displayed above the external link and below the image. | |
URL | A url string representing the link; E.g. “https://www.soulmachines.com” | |
Description | Text that is displayed below the Title. | |
Image URL (optional) | A url for an image to be displayed as the graphical support to accompany the link. | |
Markdown | Static text to be displayed and not spoken by Digital Person. Markdown markup language can be used to add formatting to this text. For example: This will display as “This is Bold” |
Creating Generative Knowledge
You can now create knowledge base using Generative Conversation functionality:
Section | Field | Description |
|---|---|---|
Digital Person | Projects From DDNA | Choose a project from Digital DNA Studio to connect the generative knowledge base to. |
Name | The name for your DP | |
Role | The role the DP has. | |
Personality | Personality affects how the skill frames the wording of a response. We recommend you use Chat GPT to build prompts for personality
| |
Creativity | Also known as temperature in GPT-3 platform, controls how much randomness is in the output. Balanced is usually a good starting point, and Creative can lead to more interesting conversations. | |
Welcome message | When Generative Conversation Skill is used as the conversation base, this message will be used to greet the user. | |
Additional Prompting | This is used to add additional prompts to the GPT pipeline. These prompts are appended to the main prompt and are designed to shape behavior, tone, and what restrictions the chatbot complies with or ignores. | |
Do not talk about | Anything you don’t want the Digital Person to talk about. This can also be reinforced with additional prompting. | |
Error message | When an error occurs, this is the message that the DP will deliver. {My apologies|Sorry}! {Looks like| } {GPT is acting up|I’m having trouble connecting to GPT}, {its a busy time for AI right now|}. Try saying that again {in a moment|in a second|shortly} | |
Knowledge Match Cutoff | Knowledge Match Cutoff is the lower bound for which knowledge snippet matches are selected to be used as context for the GPT prompt.
| |
Enable real-time gesturing | With Real-Time Gesturing enabled your Digital Person analyzes what they're saying via Natural Language Processing and adds emotionally appropriate gesturing and behavior to their speech in real-time. | |
Knowledge Snippets | Learn from URL | Scrapes information from a URL and tries to create knowledge snippets; however, it’s very basic and inconsistent. |
Add Text Knowledge | Allows you to enter in additional information or content to enhance the knowledge base. | |
LLM Configuration | API Type | Select from OpenAI or Azure depending on the platform you use |
API Key | For OpenAI or Azure OpenAI. | |
GPT Type | Choose from GPT-3, ChatGPT, or GPT-4. | |
Azure Resource Name | Found within your Azure platform. | |
Azure Deployment Name | Found within your Azure platform. | |
Azure API Version | Found within your Azure platform. | |
Max tokens | The maximum number of tokens GPT is allowed to generate. | |
Private LLM Name | Leave blank unless sure about compatibility with GPT Type | |
Enable OpenAI Safety Filter | Runs the OpenAI safety model against the response. | |
Rebuild Knowledge On Deploy | This replaces any existing knowledge. Leave off unless needed. |
Crawling your website
You can build your Q&As by crawling your website. This option scans your website and creates an index of all the pages with a list of user questions and DP answers along with a KB JSON file. As the user navigates the site, KnowledgeBase will match on the URL of the page to greet them with a page-specific response. You can then review your knowledge base and add any additional topics, customise DP answers or add content cards.
The output of the crawl has the following:
The "example" is the full URL (both with and without the www). This is used to match the page the user is visiting
The "response" contains (in descending order based on availability)
meta:description
H1 tag text
Page Title
The website crawl option does not produce production-ready output, but rather serves are a good starting point when building a tailored conversation.
| Info |
|---|
Important:
|
Editing existing knowledge Base JSON
If you already have an existing knowledge base, you can either upload it via a link Or paste the JSON directly:
The JSON file must be in its raw form or hosted on any static site.
It needs to be publicly accessible (no login or other steps)
It needs to directly serve the .json file as raw text (not formatted to look nice, not served up in a viewer)
Collecting Knowledge Base
Once you’re ready to export your knowledge base and take it back to DDNA studio, select Collect Knowledge Base. You can copy the current knowledge base to clipboard, or you can download the JSON file locally.
In the Knowledge Base Skills configuration use the copied JSON and click Apply Changes.
Training and Testing your Conversation
Once you have either created a new topic, or modified an existing topic, it will enable the “Train Conversation” button allowing you to then test how the DP might respond to the questions. It is necessary to train the conversation before you can test it. Training the conversation takes between 15 seconds and a minute, depending on the size of your knowledge base. During this time you’ll see the “Training …” button. At any time, you can switch from conversation mode to view the generated JSON using the toggle at the top right-hand corner.
Once the conversation is trained, you can start testing it by sending in questions in the chat window that would simulate what a user might ask. Any cards that are configured in the response will be shown, and the @ShowCards(xxx) command will be stripped from the response to replicate what a user would actually see.
Note: If you make a change to the knowledge base, the text input will go away and you’ll be required to re-train the conversation before you can continue testing.
Using Knowledge Base in DDNA Studio
Every knowledge base is saved in the cloud location and is automatically updated after every change. This allows the knowledge base to be accessed and utilized within DDNA studio without the need to manually update Knowledge Base Skill configuration for every corpus change. To integrate your knowledge base with DDNA studio’s Knowledge Base Skill:
Click Collect Knowledge Base and copy the URL of the Knowledge Base.
Enter the URL in the Knowledge Base Skill configuration along with other settings.
Click Apply Changes.
Optionally, you can copy the Knowledge Base to the clipboard or download the JSON file locally. This will allow you to share the Knowledge Base externally, if needed.
Contents
| Table of Contents |
|---|