Skip to main content

Core Components

Pre-Requisites​

  • Access to any Watson Assistant instance on a Lite, Plus Trial, Plus, or Enterprise plan
  • Access to an instance of Watson Discovery
  • Access to WatsonX platform
  • Access to an instance of Watson Machine Learning

Core Compontents​

  1. Setup Watson Discovery
  2. Setup WatsonX and integrate with Watson Discovery
    1. Create Watson Discovery custom extension
    2. Create WatsonX custom extension
    3. Configure WA action to integrate WatsonX Search using Watson Discovery
  3. Setup Watson Assistant
  4. WatsonX Customization
    • Add source link to response
    • Configure model response by customizing prompt

Optional Components​

  1. Integrate Watson Assistant with NeuralSeek
  2. Integrate Watson Assistant with Microsoft Teams
  3. Integrate Watson Assistant with ServiceNow

Step-by-Step Instructions​

0. Create a new Watson Assistant​

1. Setup Watson Discovery​

  1. New projects, input Project Name, select an option "None of the above — I’m working on a custom project", click "Next"
  2. select "Web Crawl", click "Next"
  3. Input Collection Name
  4. Input starting URLs, click "Add" to add. Repeat for all domains.
  5. Upper left Hamburger icon -> Manage Collections -> New collections
  6. Select data source
  7. If webcrawl, input url links to "Starting URLs" and click "Add" -> Finish

2. Setup WatsonX and integrate with Watson Discovery ​

High-level Steps​

  1. Create Watson Discovery custom extension
  2. Create WatsonX custom extension
  3. Configure WA action to integrate WatsonX Search using Watson Discovery

2.1 Create Watson Discovery custom extension​

  1. In your assistant, navigate to "Integrations" page.
  2. Click "Build custom extensions" -> click "Next" -> Input Extension name Watson Discovery -> click "Next"
  3. download json file: watson-discovery-query-openapi.json and import file to WA
  4. click "Next" -> click "Finish"
  5. Lower Right corner of the Watson Disovery extension, click "Add" -> click "Add" -> click "Next"
  6. In Authentication page, in the Authentication type dropdown, select "Basic auth"
    1. For Username enter apikey
    2. For password, create and copy a new API key from API key
    3. For discovery_url, within IBM Cloud -> resource list -> Watson Discovery Instance -> Manage -> Credentials -> URL
    4. Paste URL into discovery_url and remove https:// from the beginning of the string
  7. Click "Next", click "Finish", click "Close"

2.2 Create WatsonX custom extension ​

  1. In your assistant, navigate to Integrations page, click "Build custom extension" -> click "Next" -> Input Extension name watsonx -> click "Next" .
  2. download json file: watsonx-openapi.json and import file to WA
  3. click "Next" -> click "Finish"
  4. Lower Right corner of the watsonx extension, click "Add" -> click "Add" -> click "Next"
  5. In Authentication page, in the Authentication type dropdown, select "OAuth 2.0"
    1. For Apikey, create and copy a new API key from API key
  6. Click "Next", click "Finish", click "Close"

2.3 Configure WA action to integrate WatsonX Search using Watson Discovery ​

Upload Actions:​
  1. Download discovery-watsonx-actions.json
  2. Navigate to "Actions" page, click "Global Settings" icon on the upper right corner
  3. Navigate to Upload/Download tab, upload the downloaded JSON file discovery-watsonx-actions.json onto the tab or click to select a file from your local system, then click "Upload", and "Uplaod and replace".
  4. within the Actions page, navigate to "Actions / Variables / Created by you". Set discovery_project_id and watsonx_project_id session variable
    info

    Where to get credentials

    • discovery_project_id: within Watson Discovery: Upper left Hamburger icon -> Integrate and deploy -> API Information
    • watsonx_project_id:
      • Go to WatsonX Platform
      • Projects (click on project)-> Manage -> General -> Details -> Project ID
No action matches Setup​
  1. Navigate to "All items" -> "Set by assistant" -> "No action matches".
  2. Click on the "No action matches" action and delete the existing step 1 and step 2.
  3. "New Step". In the "And then" section, select "go to a subaction" -> select "Search" in the dropdown options -> "Apply".
  4. "Save" and "Close"
  5. You're all set. Navigate to "Preview" to test the integration!

3. Setup Watson Assistant ​

  1. Within IBM Cloud, click "Launch Watson Assistant".
  2. We could customize the Assistant Interface. In "Preview" tab -> Customize web chat.
    • Style: can change the color theme of Watson Assistant
    • Launcher: edit greeting message
    • Home screen: cusomize conversation starters
  3. We can customize the preview background to match any organization homepage by "Change background" -> "Enter Url" option -> "Continue" -> Enter website url.

4. WatsonX Customization​

  1. To add source link to the response, We need to configure two actions.
    1. Navigate to "Seach" action
      • step 5, click "set new value" within the set variable values section.
      • click "New session variable" from the dropdown, input Name source_url, select Any for Type, click "Apply".
      • in the text box after 'To', select "Expression" from dropdown, and input ${step_474_result_2}.body.results[0].metadata.source.url. Click "Apply".
      • Save, and Close.
    2. Navigate to "Generate Answer" action.
      • In step 10, in bottom of the 'Assistant says' text box, type For more information, click $source_url, Enter.
      • Save, and Close.

Filter confidence score from Watson Discovery​

  1. To access the confidence score of source link from watson discovery, we used expression such as${step_474_result_2}.body.results[0].result_metadata.confidence
  2. We recommend setting a variable confidence_score to be used as a filter
    1. Navigate to "Seach" action
      • step 5, click "set new value" within the set variable values section.
      • click "New session variable" from the dropdown, input Name confidence_score, select Number for Type, click "Apply".
      • in the text box after 'To', select "Expression" from dropdown, and input for example ${step_474_result_2}.body.results[0].result_metadata.confidence. Click "Apply".
      • Save, and Close.
    2. We could start setting conditions using confidence_score variable.

Configure model response by customizing prompt​

  1. When a user prompt Watson Assistant with keywords such as "answer in bullet points", we would like the model to output in bullet points format. This requires additional configuration in the "Generate Answer" actions.
    1. Navigate to "Generate Answer" action
      • Add a new step after step 5
      • Select "Is taken with conditions" from dropdown
      • Underneath "If All of this is true", click on most left box, and select "Expression" from dropdown
      • Enter ${query_text}.toLowerCase().contains("answer in bullet points")
      • Click "Set variable" button to the right of Is taken with conditions
      • Click "Set new value", and type model_input and select variable
      • In the box after "To", select "Expression" from dropdown.
      • Enter the follwing text
      ("<s>[INST] <<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Be brief in your answers. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n\nIf a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don’t know the answer to a question, please don’t share false information.\n<</SYS>>\n\nGenerate the next agent response in bullet points by answering the question. You are provided several documents with titles. If the answer comes from different documents please mention all possibilities and use the titles of documents to separate between topics or domains. If you cannot base your answer on the given documents, please state that you do not have an answer.\n\n").concat(${passages}).concat("\n\n").concat(${query_text}).concat("[/INST]")```
      - ![Customize WatsonX Prompt](https://github.com/ibm-client-engineering/solution-ithelpdesk-watsonx/tree/main/docs/3-Create/Customize_WatsonX_Prompt.png)

Inactivity timeout​

  • In Watson Assistant, navigate to Environments.
  • Click on the "Settings" icon next to Draft environment
  • Navigate to "Inactivity timeout" to set the timeout limit

References​