The Voice AI Platform Built for Enterprises That Actually Need to Scale

Karim Malhas: Everybody's competing to get the best agent, the best latency, the best all of these things, but people are forgetting. Once you get this, how are you going to run it?Sam Nadler: Built This Week, breaking it down. Built This Week, we show you how. A fresh idea, a clever tweak, you locked in. YouSam Nadler: Hey, everyone, and welcome to Built This Week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups. I'm Sam Nadler, cofounder here at Ryze Labs. And each and every week, I'm joined by my friend, business partner, cohost, Jordan Metzner. How are doing today, Jordan?Jordan Metzner: Yo, Sam. Happy to be back. Another big week in AI. Record fundraising again across the across the foundational labs. I mean, these guys raise money every single week.This is ridiculous. These numbers keep getting bigger and bigger. But, yeah, it's been an been another awesome week for AI, and looking forward to our guest today. Obviously, a topic I'm super excited about and really like talking about, but, yeah, looking forward to today's episode.Sam Nadler: Yeah. And I I'm super excited about our guest today. Before I introduce him, please don't forget to like and subscribe. We have new episodes of Built This Week out every Friday, and we've just surpassed the 26,000 subscriber mark on YouTube. So exciting.So let's get into the docket. We have a special guest, a CEO of Breeze, Kareem. Kareem, I think today you're gonna demo a little bit about how your platform and technology works. But first, if you don't mind, just give us a a quick intro of who you are and and what you're doing, and then we'll jump into the demo.Karim Malhas: First of all, Sam, Jordan, thank you for hosting me. I really appreciate it. And I am the founder and CEO of Breeze. I'm also the CTO. I'm a technical founder.Breeze is basically think of VAPI meets n eight n for enterprises, totally focused. This is kind of what it is. I think maybe if I start and show you a bit about it, we are totally focused on enterprises, and we're focused on partnerships mainly. And this is how we are growing, and we're scaling into multiple places in the world. We're not we're mainly focused in the MENA and South American markets and expanding into Europe and going into South America.We've launched in April, came out of stealth year of building of we we it was a tough year of building. And now we have Breeze, and it's expanding.Sam Nadler: Perfect. Let's jump into a demo and just get into the nitty gritty of how it works.Karim Malhas: So I'm gonna share my screen now. I haven't built a demo, and I'm just gonna kind of go into the platform and Perfect. Share. Basically, Breeze takes the best of every single platform. So think super base area, n eight n, VAPI, where you have, you can divide projects, you can have segregation, you can have teams, you can have monitoring, you can all of these things that enterprises really require.So this is an empty project. And you come in, you can create different project. You know, any 10 when it came out only had one project, and you have everything inside of it. I used to work in a company in a big enterprise, and that just doesn't work. So you need to be able to divide.So let's create a demo project, you can go into that. Inside your demo project, can then start creating workflows. So let's do a demo, and then it is very similar to any 10. So the concept of Breeze is simple. We're provider agnostic.We are multi agent, so we don't want to build the models ourselves, at least not today. We because we know that the models are getting better. You see the news. People are putting money and people are competing. We want to take the middle layer, which is the orchestration, and allow you to customize because, genuinely, we believe that even within the voice agents the voice agent sector, you even within the same vertical, you still need customization.You always need customization. So we built into this version control, compliance checks, and everything you can imagine. This is how you would start. And initially, we're trying to make it very intuitive. I love to build things that people are easy.So any voice what is a voice agent? It's ears, brains, and voice. So ears, you pick whatever voice, whatever thing you want. Let's go with Deepgram. You have all the models.You are familiar with Anyten. This is kind of similar, and we did it on purpose. It's all built by us, but it's similar. Let's pick models. So Celebris has a very nice model.I think the Odyssey, the GLM is a very quick one. You have the voices. We have you can integrate with any voice that you want. So eleven Labs is, like, one of the best, very fast. And then you have of course, if you go into the master, we by the way, we believe in everything being opens open, so we want people to control.You can then go. You can add any inputs. So maybe it's an outbound call, so I wanna call it with an API or anything. You can add any parameters you want. You can introduce webhooks so that you have the webhooks.These are by the way, this is all standard stuff. We've perfected the standard layer, and we've I'll tell you a bit about the second layer we're going after in a bit. It's the operational layer, I think, where we're really killing it. This is all you know, what I'm showing you now is the base layer where you build agents. So I pick a name.So I wanna have a name, maybe something else, and then you can go and you can check. Okay. You have two things that you can pick from. Okay? Either the standard path steps, so maybe I wanna do a check if the, name is equal to Kareem.I wanna maybe do something, and then maybe I wanna play some background music, office music. If it's not fire. That's a weird one. Maybe I wanna do something else. If it's if it's not Kareem, I wanna do, let's maybe make him say something or whatever you wanna do.But in this case, you can pick and choose from different path steps. You can go to a database. So let's say, if it's Karim, I wanna start a fire music, like fire background. If it's not, I don't want to, you can then go. You can go and select.This is a standard HTTP request. It allows you to access anything. So if you, like, test it, htpw.requestcatcherforce.com, for example, slash test, you can send stuff to it. We're trying to make it intuitive for people. So assume this is, going to a CRM retrieving some information, so maybe it will take the number.So I can go and be like, oh, I need the number. Which number? I want the number of the other party because it's confusing when you're doing inbound and outbound. You can say, sure. Send the number in which format, apply it, and then let's then proceed to an agent.So in both cases, I wanna go to an agent. Now agents, in this case, can inherit personas or not. So if you wanna change along the way. And this came on from the idea of lang chain and lang graph, how do you change. K?And then you want does the agent is the agent going to inherit the chat or not? If it is, is it going to know that it's a new agent or is it an updated agent? These are very minor details, but they really make a difference. So in general, the problem with single agent voice agents is that if you give it a big prompt, it knows everything. And sometimes, you don't want it to know everything.Sometimes, you need it to go through steps because I don't want you you're not supposed to skip this step. So in this idea, it's like you are a consent agent, that is your only role. Oh, maybe you are talking to Kareem, so it will know the name. You you may wanna tell it the time. The time right now is, k, the time, In which format, and in which which way are we now?We are in, like, let's say, we are in New York. Okay? So this is the time. So this is a simple agent. Okay?And now, what do we want to do? We want to go and we want to give it tools. Okay? Consent is given. And what is another tool?Rejected consent. Okay? What do I wanna do in this case? I wanna go, and I wanna say I wanna do a say, thanks. Have a good day.Goodbye. I don't wanna proceed if he doesn't say consent or he wants to end the call. Call this tool if the user rejected consent or wants to wants to end the call. By the way, this is a very, very simple demo. We've built extremely complicated things on it and just wanna, like, give you an idea of things.You can also always end the call, cold transfer, warm transfer, whatever you want. Now let's just end the call. Now what happens this is where I talk about multi agents that you control it. You multi agents so you gave consent. What do you wanna do?Oh, first of all, when I saw this agent, maybe I want him to what do I wanted to say? I wanted to say, hey, at Kaneem. This call is recorded. Do you have can we have consent to proceed? Okay.And that's it. Next, I wanna go and I wanna say, perfect. Perfect. What is your account number? Like, number.Actually, you're guiding them. This is why we need to guide people into these because when you are in an organization and building something, you always have a script. You don't, like, just to give it, like, agent total agentic. You you people are scared of full agentic. So here, your role your role is to gather the to gather the account number.K? You must repeat it digits by digits before proceeding. And maybe I I wanted to know what happened before, and I wanna give it one tool, account collected. K. Call this tool to collect the when you collect the account number.Alright? And I want to get the account number, the account number in digits. Okay? Maybe you want a list. Maybe it's multiple account numbers.Maybe it's a list of objects. And this is where we've noticed that, like, if you're building something like, food delivery, you have the menu and you have the sub items and you need I multiple of them. And you can go into this and say it's a list. It's an object, list of objects, and each one is so on. And this is we've actually done this.Now ISam Nadler: have a question. I have a question. So I can totally see how you're, like, putting, like, incredibly powerful tools into operators' hands to build their own solutions, like, similar to n eight n. How do you think, you know, one of in our, you know, brief use use case of building voice agents, latency is, like, a huge challenge. How does this either, you know it does how do you balance the quality versus latency challenge?Karim Malhas: That is an excellent question. We've actually discovered that we cannot add too many agents because every time you're making a decision to go from agent to agent, it takes time. So we've discovered that we only use these when we need control, compliance, or security. So this agent might still come in and ask you, like, imagine this. Imagine you answer the call and instantly an API sends to your database or your server, which sends you an OTP, and this agent says, hey, just to verify, please tell me your OTP.Okay? This guy doesn't know anything about the OTP. So you may tell him what is the OTP. He doesn't know anything because it's in memory. So then its only role is to collect the OTP.And when it collects the OTP, I can then take it and do an if statement, which is code, so it's deterministic. And I if it's not, I go back. Ah, you didn't match. So I don't so now we have the layer of security that enterprises require and guarantee. Because I don't trust the agent with giving it the OTP and making the comparison within the agent, cause there is always a way to jailbreak it.I studied cybersecurity and I love to, like, kind of try to break things, and that you can break. But here, you can't. So you need to balance it. Initially, when I started to build this, I thought we're gonna build everything. Now I'm you're probably gonna need two, three agents tops because an agent is super good and life, especially when you're dealing with, like, any sort of situation, it's not deterministic like this and that or this.The humans and when you're ordering, for example, I may start and make an order, then I'm going to the end, and like, I'm done, and like, oh, no. I forgot something. So if you're gonna go back and forth, back and forth, it's better to have it in a single agent. But there is situations where, like, I need consent and I cannot proceed, or I must get the OTP where this comes in handy. So this is kind of where you go.And now in terms of latency, you also have the models. This is why I love that we are agnostic and provide we don't care about we have this is a hundred milliseconds. This one is ninety milliseconds, and this is super fast. So you get the latency that you want. And I know this is kind of pivoting away from this voice agent, but it's a super important thing.And another thing is like phone numbers. We don't believe in phone numbers. We have the concept of phone groups. Phone groups, because I learned there's phone groups. The concept of phone groups that we have is that, hey, you think if you think you're going to use one number to run 10,000 calls in a day, well, you're going to be blocked.So we created the concept of, phone groups. So you you add all the numbers you want. You can add 10,000 numbers, and you then you use a phone group, and it automatically routes. We have added in it something called smart routing, which I learned from VICIDAIL. VICIDAIL, depending on your state, if I have 10,000 numbers and I call you in DC, it will use a DC number out of the 10,000 number, and it will randomly select.And these are things that are super important if you want it to work. And in Jvapi and the rest, well, it's on you to build it.Sam Nadler: You don't have to love working out. You just have to do it. HipTrain makes it easier to show up and harder to quit. Less thinking, less skipping, more consistency. Built for people who don't love fitness.Jordan Metzner: Okay. Cool. No. I mean, I think you know, I think you showed us some great tools of, like, how to build these kind of custom custom workflows for lack of a better term and kind of, you know, how you can kind of build security around that as well. You know, without kinda revealing too much, can you tell us a little bit of, like, who's the ideal customer, like, using this?Is this, like, you know, is this, like, a plumber building it for his, like, local shop? Is this, a bank building this for kind of, like, you know, getting their customers through their off funnel? Like, tell us a little bit of kind of, like, who's your, like, you know, ICP kind of, and who's found a lot of success here?Karim Malhas: That is a lovely question because the ICP is a very, very deep question that keeps evolving, but in the end, we reached a point where I think our ICP are COOs that want operational efficiencies, honestly. Because when we go to every meeting we go into, we they start off, tell us about the latency, tell us how good is the accent and the dialect, because we're working in the in The Middle East, we're working in South South America, we're working in Europe, everybody's like accent latency. I go in like we just did right now. We start to, like, talk about these things, and then I talk to them. Like, you know that your problem isn't really, like, the voice agent itself.You know, it's running the voice agent. That is actually what the problem is, really. And then you we're noticing that it's hitting so hard. People now because voice agents are so new and people are so scared of it, they just need to make sure that first it works and it's fast and good accents, and, like especially in places where dialects matter because people don't like the somebody calling them with a different dialect. And in America, standard English.It works. But if you go to any other place in the world, from country to country, it matters. And they do not like it if it's not the same dialect. So they care about that. But as soon as you do that, then you notice they notice that, okay.You've given me an AI, you've given me an API, and now you've given me a webhook. Well, how do I make it work? Oh, your development team can go. Well, my development team takes, like, three months to finish, like, a small project. How am I gonna explain to him how to build this and do it?You know? And and that's when we come in and we tell them, guys, this is what we've done and what we've noticed, by the way, is that the meeting shifts away entirely from, oh, is this does it even work? They don't even ask me. I I honestly, I've been in meetings where the beginning, it was like, I need to see it work, and then they're like, okay. Okay.Okay. Okay. Now we see what you have. And by the way, we are doing it in a way where we're partnering up because we are a small team. We are 10.I'm now going to fifteen, twenty soon. I don't wanna go. I wanna I know how the new generation of companies are. I wanna keep it small. I am in the in the I am the mentality of partnerships.I've built a system in this where the billing can be customized entirely for every organization. There's a standard. And I'm building a layer for partnerships where any partner can have multiple agreements with me, and they become responsible for onboarding clients, and they can set whatever price they want, and they upload the contract, and I approve it. That's it. So that we can scale.So I have I have partners in in in UAE, Dubai. I have partners in South America. I have partners in Kuwait. I have partners everywhere because I am willing to share this. I am building an now I'm calling it the operational layer of voice versus the infrastructure of voice.Now we are the I honestly, we have killed it in the infrastructure. Everything you've seen here from SIP trunks to, like, latency to concurrency to all of this stuff, we've we've we've nailed it, honestly. Because we are now looking at the next layer, next step. Partner oh, first one is the operational, and the second is the b to b to b and b to b. So our ideal customers are operation guys, honestly.The the guy the big the big boss in operations because he's the guy who needs to, like, make it cheaper, make it better, make it more efficient. That's what they care about in the end.Jordan Metzner: Awesome, Deno. Really cool, Kareem. And, obviously, you know, this is the edge of where, you know, business is happening today, AI voice, and obviously a lot more to come from everybody. The Foundation Labs, you know, Eleven Labs, some of the players you talked about. So really exciting space, and thanks for joining us today.I think maybe we can jump into the news, Sam.Sam Nadler: Anthropic about to raise another 30,000,000,000. I think what is shocking is the valuation at a $900,000,000,000 valuation. If I'm not mistaken, that would make it the second most valuable private company in the world, just second to SpaceX. Could be mistaken there, but that's what I quickly gathered. I think this would also surpass the OpenAI valuation.But just, you know, I think Anthropic's success throughout 2026 has just been phenomenal. And with Cowork and Cloud Code, you know, we use both over here, Kareem, Codex, and Cloud Code. But, you know, just curious what what you guys think about this this news.Karim Malhas: It's wonderful news. I love it. It's David and Goliath. I was in San Francisco in October for Tech Week, and I was standing there in I forgot which event on the one of the big towers. I was looking outside.Was like, this is the capital of the world now because new gods are being created, David and Goliath, and this proves it. This is it. Look at it. This is a company that started, I don't know, five years ago that is actually threatening Google, actually threatening them Because you don't go to Google now to ask it any question. And the the idea that Google is investing in them and Microsoft is investing in OpenAI is showing you that this is the world that is changing.And I genuinely think the next two years, new. It's a genuine change that has never happened in the history, and it is hitting every industry. This is not like like when, like, Internet came out. Oh, you can take your time. It's like if you don't do it, someone is gonna do it tomorrow, so you better do it quickly.This is the the the mentality now. It's not and, like, no. And it's not an and people are scared of AI coming and, like, taking over. No, man. It's somebody who knows how to use AI.Like, somebody who's 18 years old will come and, like, beat you. Like, I I in my company, we we must use AI. Actually, we have a competition who's gonna use cursor more. Actually, I'm still I'm I'm I'm I'm a top I'm top 8% cursor user in the world. And and I challenge my team.I'm like, come on, guys. You gotta, like, use it more. Abuse it. I have six. That's how I built it.And to be honest, the reason I'm building this company and because I want because I feel scared for the next generation. It's beautiful. For people who are in this world, it's beautiful. It's new opportunity. But people who are like, no.No. No. It's it's man, you really need to go into this world. You really must use it no matter what even if you're a photographer, you should if you are anything, anything, anything in the world, I cannot imagine something where this AI doesn't help you, make you better. But this is gonna affect the world, 80% of the world, I think, in a way that's horrible, and 20% of the world are going to go up.I think in two years, we're going to see the impact. Now do you do you hire a junior anymore? Like, ask yourself genuinely this question. Not in, like no. Not in coding.You you you catch ATPs and mid level easily. If even if you don't know how to code, it becomes mid to high level. Well, if you know how to use these tools, you becomes a superhuman. So this is what I'm trying to move and by the way, I am building this company from The Middle East, from Jordan, from around this area, specifically because I really think that I need to motivate the next generation of kids to look into these and to push, and now there's no more limits. Before, only people in San Francisco could do it because you I came to San Francisco.I went to a restaurant. Next door was fire was like what was it? Open AI on the Internet. I'm like, man, is this Open AI? I'm like, yeah.I'm like, cool. They know you can ask him. Anywhere else in the world, knowledge isn't. I built a SIP trunk. Now you know SIP trunks, how come I was scared to build.I can now build a SIP trunk with my eyes closed, and I don't fully understand it, but you can build it. You know? But what'sSam Nadler: Jordan, what are what's your thoughts? What are your what do you think about the the anthropic fundraise or the talks about the the valuation and the numbers and what it means for, you know, its competition with OpenAI?Jordan Metzner: Yeah. It's an I mean, at one point, it's kind of unheard of. Right? We've never seen private fundraising this large. I mean, I don't even know if we've seen public fundraising this large, to be honest with you.900,000,000,000, close to a trillion dollars puts it, you know, pretty much really close to the mag seven players as a company that's been around for, I don't know, just a few years. I mean, don't forget Dario worked at OpenAI, so, you know, he had to have started it after the fact. OpenAI as well getting the same type of kind of, like, record valuations. And, you know, are these companies undervalued, overvalued, properly valued? What does it, you know, what does it mean for is Facebook and all these mag seven companies?You know, if you compare the value, look how much free cash flow Facebook and Google spits off and you compare the price point. You know, maybe Google and Facebook are cheap. You know, the rest of mag seven is cheaper. Maybe these companies are expensive or, you know, maybe I just think people don't understand kind of the impact coming here. But, yeah, it just seemed like, you know, there's also this, like, Amazon owns a huge piece of anthropic.Right? And so, like, what does that mean for their value as well? Right? So a little bit of circularity as well. But, anyway, I think to Kareem's point, you know, it's good for the ecosystem.Kinda more money in the ecosystem just drives up, like, on a, you know, better product and feedback and, you know, just looks like you can kinda continue to build on top of these platforms. And it's so exciting every day. I mean, there's, like, something new, whether it's from Claude, whether it's from Anthropic, or whether it's from OpenAI, or even just, some of some other some other foundational labs, even, like, to Karim's point, the startups like his own, you know, built on top of, you know, bringing these technologies together. So, yeah, really exciting time and fun.Sam Nadler: I'll just wrap up with one news article today. Kareem, it was a pleasure. Where can people find you? Can they find you on x or, LinkedIn or your company website? What's a good place to reach out?Karim Malhas: LinkedIn, x, Instagram. I don't know. Yeah. Any of those. X LinkedIn LinkedIn, I guess, and x also as well.Yeah.Sam Nadler: Perfect. We'll we'll put those up on the screen, and, it was a pleasure to have you. Thanks for the demo. And any any final words, Jordan, before we sign off?Jordan Metzner: No. Honestly, Kareem, thanks so much for joining us. This was a really fun episode. Great to chat, see your product, and how it works. Just really quickly, Kareem, tell people, like, where to find it.Just give, like, the website spiel, like, you can find us at kind of thing. But before you do that, just yeah. Anyway, thanks thanks for joining us. That was a great episode, Kareem, thanks for having us.Karim Malhas: Thank you for having us. And, yeah, you can find us at hey breeze dot a I without an e.Sam Nadler: Thanks, guys. Bye.