SignalWire’s AI Agent is a low-code tool that allows you to integrate AI into your voice applications with ease. With options to build a virtual agent using our no-code interface or to program your own, anyone can build their own customizable AI voice assistant.

AI agents are particularly skilled when it comes to taking over customer support operations. Virtual assistants powered by AI can handle customer inquiries, provide basic troubleshooting, and perform tasks like sending messages or scheduling appointments. In this post, we'll dissect a code snippet for an AI agent designed to be a customer support agent for a cable company.

AI Zen: A digital employee at the cable company Livewire

In this example, we’ve built an AI voice agent named Zen, a witty AI assistant proficient in troubleshooting and support. Zen works for a cable company called Livewire, and is dedicated to making your experience as smooth as possible. Zen assists you with technical issues but also adds a touch of humor to every interaction.

To see how Zen works, call Livewire at +1 (201) 366-6039.

In the following snippets, we’ll break down the code behind Zen, which you can follow as an example to build your own customer support AI agent.

Server OS/applications used:

• Debian 12

• Perl

• NGINX

• SignalWire AI Agent

Below, you can see the prompt that defines Zen’s personality, behaviors, and parameters. Zen boasts an impressive set of skills, from rebooting modems to conducting speed tests and checking modem diagnostics. Zen's abilities go beyond technical tasks; he can also transfer calls and verify customer details.

The conversation with Zen follows a structured flow to ensure efficient troubleshooting and support. Beginning with customer verification, Zen requests the account number and CPNI. If unsuccessful after four attempts, Zen moves on to step two – performing a speed test and relaying the results. Steps three and four involve checking modem levels and SNR. After providing assistance, Zen concludes the call. These actions are defined in plain language:

Functions

This example requires some code in order for Zen to perform functions such as speed tests and modem swaps. Functions in this example will connect to a database and pull data with a json output.

Once the function verify_customer is executed by the user verifying the account_number and cpni then the record for that customer can be used later. The prompt for this example won't allow any further steps to happen unless verify_customer is validated.

verify_customer

• Name: verify_customer

• Purpose: Verify customer account number, cpni first, last name and phone number

• Argument: account_number|7 digit number,cpni|4 digit number

Meta data functions

meta_data functions access the output of an already executed function. In this example once the function verify_customer is executed by the user by verifying account_number and cpni, then the record for that customer is accessible via json for any meta_data functions.

modem_swap

• Name: modem_swap

• Purpose: Swap the users modem

• Argument: mac_address|new modem MAC Address in lowercase hex 12 characters

speed_test

• Name: speed_test

• Purpose: Test upload and download speed from the modem

• Argument: account_number|7 digit number,cpni|4 digit number

modem_diagnostics

• Name: modem_diagnostics

• Purpose: customer modem upstream downstream and snr levels

• Argument: account_number|7 digit number,cpni|4 digit number

Have any questions as you’re building an AI agent? Bring them to our community discord, forum, or Slack!