How to build an AI-powered Q&A bot with Pipedream

Workflows > + New Workflow

Create new Pipedream workflow

Go to pipedream.com and click the green 'New Workflow' button in the top right. You'll land on the workflow builder with an empty trigger step. This is where you'll configure the Slack mention detection that kicks off your AI bot.

1. 1Click 'New Workflow' from the dashboard
2. 2Name your workflow 'AI Q&A Bot'
3. 3Leave the default trigger step selected

Step 1 > Select app > Slack > New Mention

Add Slack mention trigger

Click the trigger dropdown and search for Slack. Select 'New Mention' as your trigger type. This fires every time someone mentions your bot in any channel or DM. You'll need to connect your Slack workspace and choose which bot user to monitor for mentions.

1. 1Type 'Slack' in the app search box
2. 2Click 'Slack' from the results
3. 3Select 'New Mention' trigger type
4. 4Click 'Connect Account' to authorize Slack

Slack Trigger > Configuration

Configure bot mention settings

Select your bot user from the dropdown - if you don't have one, create it in Slack's app settings first. Set the trigger to fire on mentions in any channel. Test the trigger by mentioning your bot in Slack and clicking 'Generate Test Event' to confirm Pipedream receives the message data.

1. 1Select your bot user from the 'Bot User' dropdown
2. 2Leave 'Channel' set to 'Any Channel'
3. 3Click 'Generate Test Event'
4. 4Go mention your bot in Slack with a test question

Step 2 > + Add Step > OpenAI > Chat

Add GPT-4 processing step

Click the + button below your trigger to add a new step. Search for OpenAI and select it. Choose 'Chat' as the action type since you're sending conversational questions to GPT-4. This step will take the mention text and send it to OpenAI's API.

1. 1Click the + button below Step 1
2. 2Search for 'OpenAI' in the app list
3. 3Select 'Chat' action
4. 4Click 'Connect Account' and enter your OpenAI API key

OpenAI Step > Configuration

Configure GPT-4 model settings

Set the model to 'gpt-4' for best results. In the messages array, create a system message that defines your bot's role and a user message containing the actual question. Map the question text from the Slack trigger data using Pipedream's step picker.

1. 1Select 'gpt-4' from the Model dropdown
2. 2Set Max Tokens to 500 to control response length
3. 3Add system message: 'You are a helpful technical assistant'
4. 4Add user message and map it to steps.trigger.event.text

Between Steps > + Code (Node.js)

Clean up the question text

Add a Code step between Slack and OpenAI to strip out the bot mention from the question. Raw Slack mentions include '@botname' at the start, which you don't want to send to GPT-4. This step extracts just the actual question text using JavaScript string manipulation.

1. 1Click + between your Slack and OpenAI steps
2. 2Select 'Code' then 'Run Node.js code'
3. 3Paste the question parsing code
4. 4Update the OpenAI step to use this cleaned text

This code step cleans the question text by removing bot mentions and extracting just the user's actual question. Paste it in a Node.js code step between your Slack trigger and OpenAI step.

export default defineComponent({
  async run({ steps, $ }) {
    const originalText = steps.trigger.event.text;

export default defineComponent({
  async run({ steps, $ }) {
    const originalText = steps.trigger.event.text;
    const botUserId = steps.trigger.event.bot_id || '<@U1234567890>';
    
    // Remove bot mention and clean up the question
    let cleanQuestion = originalText
      .replace(/<@[UW][A-Z0-9]+>/g, '') // Remove all user mentions
      .replace(/^\s+|\s+$/g, '') // Trim whitespace
      .replace(/\s+/g, ' '); // Normalize spaces
    
    // Check if there's actually a question
    if (!cleanQuestion || cleanQuestion.length < 3) {
      cleanQuestion = "Please ask me a specific question and I'll help you find an answer.";
      return {
        question: cleanQuestion,
        shouldRespond: false
      };
    }
    
    // Add context about the channel for better responses
    const channelName = steps.trigger.event.channel_name || 'general';
    const contextualQuestion = `In a ${channelName} channel discussion: ${cleanQuestion}`;
    
    return {
      question: contextualQuestion,
      originalQuestion: cleanQuestion,
      shouldRespond: true,
      channelContext: channelName
    };
  }
});

Step 4 > + Add Step > Slack > Send Message to Channel

Add Slack thread reply step

Add another step and select Slack again. Choose 'Send Message to Channel' as the action. Configure it to reply in a thread by setting the thread_ts to the original message timestamp. This keeps the AI response organized under the original question.

1. 1Click + to add a final step
2. 2Select Slack > Send Message to Channel
3. 3Map Channel to steps.trigger.event.channel
4. 4Map Text to the GPT response from OpenAI step
5. 5Set Thread TS to steps.trigger.event.ts

Workflow > Deploy > Test in Slack

Test the complete workflow

Click 'Deploy' to activate your workflow, then test it end-to-end. Go to your Slack workspace and mention your bot with a real question. Watch for the threaded AI response to appear within 5-10 seconds. Check Pipedream's execution logs if anything fails.

1. 1Click the green 'Deploy' button at the top
2. 2Go to Slack and mention your bot: '@aibot What is React?'
3. 3Wait 5-10 seconds for the threaded response
4. 4Check execution logs in Pipedream if no response appears