Introduction

Hi everyone, I'm Mario Miletta, I'm the head of product of DeepAgent, and today, this night, we are talking about AI's voice agent.

Who We Are: DeepAgent

Who we are?

We are DeepAgent, a company based in Rome, and basically we are building the lovable, as Giovanni said, but not for building apps, but for building AI voice agents.

At the moment, we are working in Italy and in Spain, but we are growing faster in the enterprise and small business sectors. In three different sectors, healthcare, telco and utilities.

From Chatbots to AI Voice Agents

and now okay I'm talking about what is an AI voice agent which is the different than the chatbot and then we will see how to build an AI voice agent with

DeepAgent now live so what is an AI agent first of all do you know what is an AI agent, the difference with a chatbot.

Chatbot vs. AI Agent: The Core Difference

Basically, a chatbot is an AI that can send output when we send an input. So, it's a text AI that gives us a reply based on what we say and what we want, like chat GPT, cloud, et cetera, et cetera.

An AI agent can do things. This is the main difference and this is why the AI agent era is something that will change the world, I think, or I hope for us.

And the AI voice agent is the area where an AI can do and can talk with us. And this can change, I think, everything, because our main interface with other people is talking not chatting and if we can talk with an AI voice agent we can

basically do everything so our goal is to build this kind of bot that maybe we can he will take our job but I don't know so okay so what is an AI voice

How an AI Voice Agent Works

voice agent is a mix of technology. We talked about ChatGVT, that is a large language model. In an AI voice agent, we use three different models.

So first of all, the user speaks with an AI voice agent. And in the AI voice agent, we have three models.

The Three-Model Stack: STT, LLM, and TTS

models, the speech -to -text models that can capture this voice and translate it into text that large language models can receive and think about in order to produce an output.

This output is taken by the text -to -speech model model that takes this text and makes an audio, a voice that will be sent to the user in order to speak basically.

And this in under one second, in basically 300 milliseconds in order to have a good latency and make the conversation natural. natural.

Latency and Natural Conversation

So if you talk with an IVR, you can send a robotic voice, the high latency.

What we are doing is to make an AI voice agent that sounds really, really natural and actualize a human.

Making Voice Agents Useful in the Real World

In fact, this is the real problem for AI voice agent. The part where you can can talk with an AI is the easiest part.

The hardest part is to make it work like a human.

This means create tickets, book a calendar meeting, update on CRM, and so on.

Orchestration: Tools, Workflows, and Control

To do this, we are working a lot on the orchestration part of the agent.

We are building the harness that can can make us to control the agent in a way that you can use it to manage different kinds of use

Key Use Cases

Inbound Voice Agents (Support and CRM-Connected Calls)

case, like inbound agent, where a voice agent can read and update data from your CRM or database. For example, if you have Salesforce or HubSpot, the agent can read this data, it can act on this data,

in order to reply in the best ways possible to the customers, in order to make campaign for the customers, in order to update the customers details and so on.

I don't know, if you call for a customer support ticket, an agent can understand who you are, what's your data, so what's your business, business, how much you pay and so on and based on the conversation it can update the information about you or it can transfer the call to other human or AI voice agent.

Outbound Voice Agents (Proactive Calling at Scale)

The second use case is the outbound agent, so when you receive a call and the call is is started by an AI voice agent.

So I don't know, for example, when Amazon call you for trading a face, this is an example. But in this case, you can use this AI voice agent

to call thousands of contacts in order to make a promotion or everything you want. And we have managed 1 million calls this year.

and the performance is really really good maybe better than the human performance so for conversion rate and so on because obviously the AI voice agent is really expert because you can train it on your data on your knowledge base with the best practice about sales appointment setting and so on so it's

Building and Deploying with DeepAgent

really the expert if you want to try the pageant this is the QR code you can try it free for 60 minutes and it's like lovable so it's just chatting with an AI that build for you an AI voice agent so it's really easy try it and now the live okay so this is deep agent so what

kind of AI voice agent we can build for example an inbound agent that can receive call and can reply to this call about I don't know so okay as you can say as you can see is like observable so there is a chat interface where you can say what you want to achieve and an AI behind the the front -end can ask you question I can

understand what you want because he's trained on the best practice to build a voice agent and in this way everyone can build an effective AI voice agent that can really work in production.

That is our main goal because there are a lot of players that can build this kind of agent but you have to try it in production, that is the hardest part. So obviously he's smarter than me so you can understand everything.

Multilingual Support

The agent can talk in a lot of languages and he can use multi -linguages when he talks. So if you have to manage customer support for a lot of countries, the agent is perfect because he can talk in English, Italian, Spanish and so on. In English better than me, obviously.

Configuration and Integrations

And you can set up everything about it. So the tone, what he have to say, the tool he have to use, like book meetings, like transfer call to human, like, I don't know, API call to other tools, because obviously you can, we have native integration, but if you want, you can add your tools like with API integration.

and in this setup he follows the best practice to build an agent and he will build this agent we have no time so so this is the outcome and obviously you can talk with the agents or call

Live Demo: A Sample Conversation

it hi olivia how's going thank you for asking me i am doing well and before we continue i would like to let you know this call will be recorded as audio transcript and summary do you consent yes wonderful what brought to deep agent today or what challenge are you looking to solve olivia you

have a beautiful voice and to make sure i got you to the right setup what kind of calls are you OK.

How to Try It

So in this way, you can follow what you said. And this demo is set up for the onboarding for the page. And if you want, you can call this number to test it.

So maybe later, take a photo. OK. Very interesting.

Closing

So, please, provide more pills for Mario.