Updated: 2025.06.06

New way to implement the workflows in the app:

• Set up a Global Prompt:

  • 
  • For example “You are a voice assistant helping with creating new AI interviewers. Your task is to collect data from the user. Remember that this is a voice conversation - do not use any special characters.”

• Modify the first Conversation Node:

  • 
  • Add a prompt like “Greet the user. Inform them that you will get some information from them, to create a perfect interview. Ask the caller for data required to extract. Ask the questions one by one, and await an answer.”
  • On the right-hand menu click on “Extract Variables” tab, and add the variables, just like in the Gather node in the video.
  • level - The job experience level. - string - you can add enum values to make sure the data stays consistent
  • amount - How many questions would you like to generate? - number
  • techstack - A list of technologies to cover during the job interview. For example, React, Next.js, Express.js, Node and so on… - string
  • role - What role should would you like to train for? For example Frontend, Backend, Fullstack, Design, UX? - string
  • type - What type of the interview should it be? - string - you can provide enums here too

• Modify the condition bubble to say something like “If user provided all the variables.” - you can click on the yellow bubble or the green edge to get it's condition in the right-side menu.

• Add a new node - API request, and configure it exactly the same as shown in the video. Remember about adding the userid variable along the ones extracted from the conversation!

• You can also configure custom messages for the current API Call status - request sent, recievied success response, and request failed:

• Add a Hang-Up node below.

• The final Workflow should look like this:

  • 

• Save the Workflow, and copy it's ID. Paste it in the .env.local as NEXT_PUBLIC_VAPI_WORKFLOW_ID - same as in the video.

  • 

• Later in the course, when implementing vapi.start() method, instead of passing workflow's ID to the method as the first variable, we have to modify it a bit. vapi.start() checks if Agent was defined, and then falls back to Workflow if necessary, so we have to explicitly say that Agent is undefined - here is the updated handleCall method:

const handleCall = async () => {
    setCallStatus(CallStatus.CONNECTING);

    if (type === "generate") {
      await vapi.start(
        undefined,
        undefined,
        undefined,
        process.env.NEXT_PUBLIC_VAPI_WORKFLOW_ID!,
        {
          variableValues: {
            username: userName,
            userid: userId,
          },
        }
      );
    } else {
      let formattedQuestions = "";
      if (questions) {
        formattedQuestions = questions
          .map((question) => `- ${question}`)
          .join("\n");
      }

      await vapi.start(interviewer, {
        variableValues: {
          questions: formattedQuestions,
        },
      });
    }
  };

• You can find all these changes in the source code provided in the Video Kit for the course.

0 Comments

Top comments

No Comments Yet

Be the first to share your thoughts and start the conversation.