Video summary

Voice Ai,

Main summary

Key takeaways

Technology

Tech concept / product tutorial: Building a Voice Appointment Assistant (Bepi/Voyce AI + webhooks)

The video is a module/tutorial demonstrating how to build a voice assistant (e.g., “Bepi / Voice assistant”) that can book appointments.

The assistant’s appointment logic is presented as a data-manipulation pipeline:

  • Collect user information verbally
  • Call a backend tool via webhook
  • Store the appointment in a sheet/database
  • Return a success confirmation to the user

It also notes that for production you may use paid APIs, but for learning/testing you can rely on free options or AI services.

Key services/tools mentioned

  • 11Labs.io: referenced as a way to create/connect voice assistants.
  • Bepi (Beb/i): the main platform shown for building the voice agent.
  • Webhook / “webbook”: used to connect the voice assistant to backend logic.
  • AD server / AT server / workflow: the tutorial references creating a new workflow (e.g., CreateAppointment) using a HTTP POST method.
  • Assistant tools: the assistant calls “tools” to execute appointment actions.
  • Google Sheets (“appointment sheet”): used to save appointment records.

Workflow setup (high level)

  1. Create a workflow in the server/backend (e.g., CreateAppointment)

    • Configure an HTTP POST webhook URL
    • Use a production URL; a test URL may be used temporarily during development

  2. In Bepi, create a new assistant

    • Example: “Advantic Labs Assistant”

  3. Configure the assistant system prompt

    • Define the scheduling context (e.g., “schedule appointments for Adventic Labs”)
    • Specify the required verbally-collected fields:
      • Full name
      • Phone number
      • Company name
      • Preferred date
      • Preferred time
    • Include guidance for optional description fields (if applicable)

  4. In the assistant Tools section, create a custom tool

    • Example tool name: “Create Appointment”
    • Tool behavior:
      • Post the collected data to the backend webhook URL
      • If successful, return a success message to the user

Assistant behavior and voice configuration

Conversation flow (testing)

  • The assistant greets the user and asks whether they want an appointment.
  • It then collects required details.
  • It calls the server tool/webhook.
  • Finally, it confirms that the appointment was successfully scheduled (often including the user’s name).

Model selection

  • The tutorial mentions selecting a model (e.g., “4.0”) for improved responses compared to other versions.

Voice configuration

  • Multiple voice options are available (speaker names like “Kelly”, “Harry”, etc. mentioned).
  • Transcription/language handling:
    • Attempts languages in a sequence (e.g., Bengali/English) and falls back to English
  • Background sound can be enabled.

Backend/tool options demonstrated (extensible)

The tutorial emphasizes that you can add more tools beyond just “Create Appointment,” such as:

  • Check availability
  • Cancel check
  • Other scheduling-related actions

It also mentions phone-call related behavior via “predefined functions,” including:

  • Enable and End Call (auto-disconnect after silence)
  • A possible dial keypad / callback concept (calling a number via phone)

Implementation details: troubleshooting / adjustments

Fixing the tool call

  • Initially, the tool call may fail because of formatting issues.
  • The fix involves correcting the path/URL parameter formatting in the tool configuration.

After the fix

  • The assistant successfully collects data and calls the webhook.
  • The appointment appears in the webhook response and/or the sheet.

Data formatting improvement

  • The tutorial then improves message quality by manipulating/formatting values in a code node (JavaScript):
    • Extract fields from the webhook request body (name/phone/company/date-time)
    • Rebuild a more polished success confirmation for the user

Google Sheets / data mapping

  • The assistant flow is connected to a Google Sheet (e.g., “Appointment sheet”).
  • Columns mentioned:
    • Name, Phone, Company, Date, Time
  • The setup uses nodes to append/update rows by mapping assistant output fields into sheet columns.

Output routing

  • The video uses a “Response Webhook/Response” node concept to send tool/assistant output back to the calling frontend (or webhook consumer).
  • It also mentions integrating the assistant into a website by calling the assistant API from the site.

Main speakers / sources

  • Primary source/platform demonstrated: Bepi (voice assistant builder and its assistant/tools/workflow UI)
  • Referenced voice/AI source: 11Labs.io (mentioned, not the main implementation)
  • Human speaker name: not clearly provided in the subtitles; the tutorial appears to be narrated by the video instructor/narrator.

Original video