Designing the API for building Voice Assistants, with Nikhil Gupta from Vapi

Mike sat down for a chat with Nikhil Gupta, founder of Vapi, to talk about designing and building a product around Voice AI for developers. They chat about the right type of onboarding for voice assistants, building tools that API developers can test easily, and the future of Voice Assistants.

Designing the API for building Voice Assistants, with Nikhil Gupta from Vapi


  • 00:00 Introduction and Guest Introduction
  • 00:52 The Future of Voice AI
  • 01:56 Wapi's Elevator Pitch
  • 03:27 Challenges and Opportunities in Voice AI
  • 05:11 Building Voice AI Solutions
  • 11:00 The Genesis of Wapi
  • 16:39 API Design and Developer Experience
  • 21:23 Docker vs Kubernetes: A Developer's Perspective
  • 22:28 Hello World with Wapi: A Hands-On Experience
  • 23:34 The Developer Experience: Making APIs Accessible
  • 26:09 The Future of Dev Tooling with AI
  • 28:43 Use Cases and Real-World Applications
  • 33:23 Technical Deep Dive: Building with Node and Kubernetes
  • 37:59 Team and Future Plans
  • 39:15 Conclusion and Final Thoughts

Show notes - call bots to practice selling



## Introduction and Guest Introduction
[00:00:00] **Mike Bifulco:** Welcome back to APIs you won't hate. My name is Mike Biko, one of the co-hosts of the show, and I am sitting down for an interview this afternoon with someone who I'm really, really excited to chat about. You, you may know if you've been listening to the show long enough for hearing me p along on the internet for long enough that I.
[00:00:15] Worked for a while on the Google Assistant team in the world of voice tech and voice software and the land of assistance and iot and automation, all that stuff. That is one of the reasons that I'm super excited to, to chat with this founder. And the other reason, of course is that Nikhil Gupta is joining me from Wapi.
[00:00:31] He's. Also a Y Combinator founder. So we're, we're we have many sort of related parallel lives seemingly going on here. I'm super excited to talk about what you're building. The pitch on the website, the, the H one on your site says Voice AI for developers, which I think paints a lot of story for a lot of people listening.
[00:00:46] But Nikhil, thanks a ton for joining. I appreciate you being here today. How are you doing?
[00:00:50] **Nikhil Gupta:** Thanks for having me, Mike. Excited

## The Future of Voice AI
[00:00:52] **Nikhil Gupta:** to chat about Wapi and so curious to hear kind of the stories of voices like Google Assistant and what's happening [00:01:00] there Now, everyone but everyone is wondering now too, if you have any needs to share with us that would be great. And it's super fascinating the, the whole world around us, just like. It's gonna change, right? Like the way we talk to com, like the way we interact with computers, the, maybe we interact with like microwaves. I think that's all about a change in the next like five years. I like the fabric of like interaction. And I think that was pretty clear to people when they saw that G four oh demo. So it
[00:01:26] decided to get into it. Yeah.
[00:01:28] **Mike Bifulco:** of course. Yeah. So we're, we're a couple weeks out, out from GBT four Oh which is the O stands for O Omni. Channel Omnis Omni. Yeah, just omni in general. So this is the first like multimodal GPT from OpenAI which I think has really changed the way a lot of people are thinking about the software.
[00:01:45] And, and so we're definitely in a changing landscape, especially since, so I, I left Google Assistant two years ago at this point. And at that time things were very different. But before we get into that, let's start with you here.

## Wapi's Elevator Pitch
[00:01:56] **Mike Bifulco:** Give, give me first of all the elevator pitch for, for Wapi. What's [00:02:00] the, the, the story?
[00:02:00] What do you sell it to people as?
[00:02:01] **Nikhil Gupta:** Yeah, sure. The elevator pitch is, yeah, voice effort, developers. The you wanna add voice spots to your phones, website apps, we provide like a modular interface that's easy to use, easy to configure, and get you up and running very quickly. The, the, the, I guess like the, the, the favorite thing for developers is like the modularity of it. You don't like open ai, you wanna use Entropic, you can use Entropic. You don't like 11 Labs. You can use play hd. There's like different options for like transcribers hey, you want to like integrate, like make Zapier, like tools, like multi assistant, like all that kind of configurability that I think exists a lot out there for like chat you can bring to voice. And the future I guess, is pretty clear to people, right? When you call into a restaurant, most likely you'll be talking to an ai, and that's kind of what we're building. The, the, the bricks and pieces the [00:03:00] shovels to help you get that up and running in a day, in a couple hours instead of like months.
[00:03:06] **Mike Bifulco:** sure. Something that would've sounded like an absolute miracle just a few years ago, maybe even a few months ago. If you think about it,
[00:03:12] Definitely compelling for, for anyone working in a world where your end users are especially consumer focused. I'd imagine there's lots, lots to be said there, but in service industry, that's really interesting.
[00:03:22] And customer service of any sort myriad use cases for that. That's, that's super cool.

## Challenges and Opportunities in Voice AI
[00:03:27] **Nikhil Gupta:** I mean, I think one thing I want to highlight there is like, I think everyone dreads like calling into United, right? Like, like at t, like tax like IRS, but I, it's pretty conceivable that we would love calling them in the future. Because they, they'll be like, they have like such great ai, they're so patient, they have so knowledgeable and you just call them in and like they can immediately answer like, what's happening with their flight?
[00:03:51] Like, can they like cancel, move it around. So there is an interesting kind of change there too. Like one, the experience improves, but does it also [00:04:00] mean like the, the world went from like phone calling to like chat
[00:04:03] because like these, it, it's very expensive for these companies to like host voice experiences because it, it involves humans in the background where it's like chat could be more structured. And now it's like what is the inherent desire for people? They wanna do, use a voice or chat. But that I think is also gonna be interesting in the future where you can just call into at t and it'll be like a great experience. Which is weird to think, but it.
[00:04:29] **Mike Bifulco:** Sure. Yeah. I love the one word you used there, which is really interesting to me, is patience. Like having an operator on the phone who's patient is kind of also a pretty big game changer. I, I don't think people think of it this way, but usually if you call in customer support somewhere, or even if it's a restaurant for example, you call in and speak to someone.
[00:04:46] They're a complicated person with a busy day and probably lots going on, and you're almost certainly not the first chat they've had that day and. I would imagine many customer service people have had negative experiences just about every day with someone who's mad about something. And the nice thing about [00:05:00] building a voice agent programmatically is that they don't get tired.
[00:05:03] They might not experience that, and if they're kind and polite and all that, it really changes the way that people interface with things. That's super interesting.
[00:05:09] **Nikhil Gupta:** I, yeah. Yeah.

## Building Voice AI Solutions
[00:05:11] **Nikhil Gupta:** And maybe I can talk a bit about, like, you know, right now, if you were to kind of try to create this future maybe you wanna build a company, maybe you wanna build, like a small agency or maybe you're a small business, like there's kind of different personas here of like who would, who would wanna use, like why ai? There's a couple options, right? Like you can take 11 labs deep Graham and open AI off the shelf and like try to switch them together. That process will take you at least a couple weeks, like if not months. Once you have stitch it together you would have to kind of figure out how to like scale that system because these are what's interesting about like multimodal and like, I guess voice specifically as soon as you enter like rounds outside our text is these are stateful long lived jobs and maybe this
[00:05:58] technical center, but. [00:06:00] When you hear stateful long lived, I think that's like alarm bells go off for people. It's like, oh, that sucks. Like, it's like, but that's the reality, like where these GPUs are running, which contain the context of the conversation. So ultimately when a person is calling, they have to be pinned to something like a resource in the backend.
[00:06:20] So scaling that out as a system is like a very complex undertaking, and that's kind of the value we bring in where if you harken back to like. Twilio or Stripe, right? Like early days payments could be done where you just had to fill like this ENT form with like Wells Fargo and like you could get set up like to accept payments on your website. Same for like Twilio, like if you wanted to send a text, like it's totally possible you could. Had a contract with like Verizon at and t. But Twilio just offered like, here's an API give me the number you wanna call, gimme the text you wanna send them done.
[00:06:55] So that kind of ease of use is what we are trying to bring to like [00:07:00] voice AI world where here's the number I wanna call, or here's my number and here's the position prompt I wanna use. And that's it. Done.
[00:07:09] **Mike Bifulco:** I, what I love about this is that it represents such a big change for, for this world. So clearly I'm gonna go pretty, pretty deep and nerdy on the, the things that have changed in voice tech in the past few years. But thinking back to the dawn of the first voice assistants the I, I'm reticent to say their names on a recording because I'll kick off actions on people's phones around the world.
[00:07:29] But you know, Apple's, assistant and Amazon's and, and Google's are all the initial programming interfaces for those were almost quite literally like, imagine a tree structure for the types of conversations you want to have and build that tree for everything anyone could possibly say. And have escape hatches for everything.
[00:07:46] And that's changed so dramatically by now that AI is way, way better at picking up with the context of the conversation and inferring things. And even like if you stutter into an ai, it won't freak out. It can understand that as well. It's, it's super interesting and you [00:08:00] really are at the dawn of something where people can build very compelling things without a lot of work without having to catalog the whole conversation.
[00:08:07] **Nikhil Gupta:** Yep. Yep.
[00:08:08] It's exactly like, that's, I think, the common thing we hear. So there's it, it's, it's interesting because like when we started Vai about a year ago, we were doing something else before, and when we started Vai, we were expecting that there's a new platform being built kind of the same way that iPhone kind of ushered in this like apps era.
[00:08:27] And then no one predicted like Uber would be the thing that like. Was a killer app. We were expecting like voices capability, that thing unlocked as a platform. There would be like some really cool apps. And I still believe that, you know, like there'll be completely crazy apps, infrastructure like applications out there that are voice experience based.
[00:08:46] And I think we are seeing that in learning a little bit. They're, they're hitting multimillionaire or I know they're hitting like. They're on track to hit like, you know, a billion dollar companies or multi-billion dollar, even ars. But that was like our kind of [00:09:00] pieces where we wanted to. I'll be this kind of infrastructure for this new platform where a lot of like new experiences are gonna be formed around voice. But the, to tie in what you were saying, oh, a lot of use cases and surprising amount of like volume, like an insanely large volume exists in like the old world of telephony. And. In like systems that people are built, like people have like teams of like 5,000, like they're designing like these like flows, these conversations. And clearly that's like shifting to like prompt, like here's the system prompt. And making better bots faster is like a, is a thing that resonates with like this crowd of like, telephony, which we are seeing a lot of adoption is I, I feel like, I don't know, I'm like hopeful. I really want, like, new experiences, new interfaces.
[00:09:51] That's why we have like, for example, like a Python, SDK, but you can like, set up like a AI toy with, it's not used
[00:09:58] as much as like [00:10:00] the polyphony stuff, but like I would love for that stuff to come. You know, as things happen.
[00:10:06] **Mike Bifulco:** Some of this I feel like people need to, I. Come to the understanding that there is this like creative space where, where you can build things. Now that used to be much harder to do. And the fact that someone could pick up a Python, SDK and in a few hours build a pretty impressive demo is like not necessarily obvious.
[00:10:23] Even if you've used chat GPTs, you know, new, new audio features in the interactivity built into that from their app, it, it still feels kind of like magic is happening on your device. It's, I feel like it's that new, you know.
[00:10:34] **Nikhil Gupta:** it could be, it could be, it could be the, the, just the newness of it. I also wonder like how much of it is like there's a bit of like stasis, I think because open air is moving so fast
[00:10:42] that people are like not sure what to build because AJ is around the corner, right? Like, what would be, what even is the point of like building anything.
[00:10:49] It's totally fair. But like at the same time, everyone knows there's like so many huge opportunities. I'm just like not seeing them being capitalized as I guess as much which is, which is interesting.
[00:10:59] **Mike Bifulco:** let's [00:11:00] take a step back.

## The Genesis of Wapi
[00:11:00] **Mike Bifulco:** You mentioned a minute ago that, that you're only about a year old into building this product, or you've only been building it for about a year. Can you tell me about how Vapi got started, like what you were doing beforehand? Where and, and what was the genesis of all this?
[00:11:13] **Nikhil Gupta:** Yeah. Yeah. So we were building ai, meaning note taker and it was called Silver Power. It was great. Hit profitability and we kind of realized the next milestone for us was like to get it to like 10 mil error, which is the usual, you know, milestone when you hit that kind of number. And we just, like, we didn't wanna spend our twenties like selling meeting software. It
[00:11:37] was. Not the most challenging not the most like forward-looking impactful thing we could be doing. We just felt like there's like a bigger technical challenge we could take on and that like we were engineers and like, we just love technical problems. So started stumbling around on like finding what we could do. The, this my co-founder one day in his. [00:12:00] Depression of like pilot pivot, depression made this thing because we over like like over the course of like our startup, we have spent around like 30 to 50 KI think on like coaching, like founder coaching. We
[00:12:13] started at the company when we were like right outta college immature. Didn't know how to work with emotions, communicate the, the usual thing of like founder complex and worked with a coach who was saved our company. Like he helped us articulate what we need from each other and
[00:12:32] worked through ourselves and our own mind, and that that coaching was transformative. So transformative.
[00:12:39] We were like, oh, that, like when, so my co-founder Jordan, he created that same experience basically with an ai where that AI could like help him. Reflect back his thoughts, his emotions, kinda like a Yeah. Therapist.
[00:12:52] The experience got better and better. But ultimately we realized there's still like, like the company that's focusing, that's focusing on like, [00:13:00] making this experience better. It needs to be that needs to be the company itself. Like we cannot take out the problem of like making a great voice experience and making it like a great therapy. But, those have to be separate companies. And we are like, oh, that's actually kinda great because we are, we love technical, we are technical, we love technical problems, we love technical customers. Like we, we just wanna work with startups and like really technical people. That it was just like a perfect alignment where oh yeah, like this is, this feels right, like making APIs that feels right for people.
[00:13:32] And that's how we got into it.
[00:13:34] **Mike Bifulco:** sure. Yeah. I think, I think especially if you've had some coaching around like trying to reach scale and get to a point where if, if at that point you had taken on investment, like you really need to tackle some gargantuan problems. And to your point before about not wanting to spend your twenties selling meeting software like.
[00:13:52] You would need to sell all of the meeting software to really hit the scale that a lot of investors are looking for. And that has happened, right? Zoom certainly had a, a bit of a rocket [00:14:00] launch over the past few years, but you know, how many times can that happen? And almost the meta problem here is, is also really interesting too.
[00:14:06] Yeah, that, that's really fantastic. I.
[00:14:08] **Nikhil Gupta:** Yeah, you're right. Like there's like, you know, something about like the space you're in which limits, like, what you can do as a feeling. I, I do find like, I, I, I go between my, in my mind between, like, I think the, the tam is never the market. The tam is the founder, which is like,
[00:14:23] but you wanna just keep finding new ways to like, go, like expand Tesla, you know, like robotics now it's like, it says random as hell. Like SpaceX became starlink. Cool. So I, I do find like the tan isol as a founder, but yeah, like focusing, like for us it was just like, we wanna spend our time solving hard technical problems and offering great experiences for developers.
[00:14:45] **Mike Bifulco:** There's a lot of wisdom in your words. It sounds like you, you've probably been through, like ingested the coaching in a way that was really healthy and have lived through a, a hard pivot with another founder. Having been down this. Startup wrote a few times myself. I've had the good, bad and middle experiences [00:15:00] of, of you know, many lifetimes I feel like wrapped up in building startups.
[00:15:03] It's always interesting to hear someone who, who acknowledges that challenge as part of building the, the company too. That's, that is a big thing that we don't talk about a lot in general, I think.
[00:15:13] **Nikhil Gupta:** Oh yeah. I mean the, we've been trying for like four and a half years now. Like at it,
[00:15:20] We were Winter 21.
[00:15:21] **Mike Bifulco:** Good for you.
[00:15:23] **Nikhil Gupta:** It is kind of, I, I, I mean, I think there's something to, like, something to, like, I think PJ and Michael, lets to say this where pg where it's like what sometimes like founders are founders because they don't have anything else to do.
[00:15:38] Like I, I don't know what else I would do.
[00:15:41] **Mike Bifulco:** Yeah. Agreed.
[00:15:42] **Nikhil Gupta:** yeah, so, so we just gotta figure it out. That's, we
[00:15:47] had another choice but to figure it out and that's where it's been a lot of fun. So I think for people, I guess kind of wondering, like develop, I'm guessing obviously very technical, wondering like what to do.
[00:15:58] I, it is, [00:16:00] I guess like the, the, the seed, the seed of like, that feels so good. I know I'm spending my time serving people I like a lot.
[00:16:11] And that is enjoyable in itself, like no matter what happens to Wapi. So that like the feed of advice that people give around, do keep doing the things you really like, that you feel like especially passionate about.
[00:16:20] And like that the moat or competitive advantage will come from that because you
[00:16:25] just have like so much experience, like so much love for that thing. So much love and serving that thing.
[00:16:30] So that's been
[00:16:31] **Mike Bifulco:** It has to be a, a problem you can care about for 10 years, you know, not just a, a 18 months or something like
[00:16:36] **Nikhil Gupta:** yeah, a hundred percent. And like.

## API Design and Developer Experience
[00:16:39] **Nikhil Gupta:** Like the, when I think about like, the common thing I hear, and maybe this could be a bubble, is that people love our API. It's like people
[00:16:47] love, like the design of it. It's like it's easy to use. It makes sense. The validation is great. And that's just like, I'm like really in know about like the little details where like it
[00:16:57] has to be like, for example, like. [00:17:00] Singular, not plural. Like it has to be like consistent in this format. There has to be like the same crud interface to all of these things. It has to be a noun. Like all these best practices that I'm sure people that I'm talking to here know about. But it's like, you know, it's surprising to me when I look at other APIs, it's like, just like, what was the design process here?
[00:17:18] Like, what the fuck was going on in that head? Like making this interface like, like makes no sense to me. Where. Having this opinion of like a good developer experience, like really caring about that gives me joy. So that
[00:17:32] I've been a lot of fun
[00:17:33] building.
[00:17:34] **Mike Bifulco:** you and I are just meeting for the first time, but I came into the world of APIs by way of studying user user experience and human computer interaction. And so being, being deeply embedded in the world of design, one of the things that I often say to people is that learning design is interesting because anyone can learn it.
[00:17:48] It's not something you're born with,
[00:17:50] but the curse of it is as you learn these rules and as you learn the, the. Things that break the rules. You see it everywhere, right? So like by focusing on building an API that follows all these [00:18:00] rules and uses plurals correctly and is thoughtful about the way interfaces are presented, you start to see all of the inadequacies of your own product, also other products.
[00:18:07] And they become more and more like little red irritations across your life. And,
[00:18:12] **Nikhil Gupta:** Yeah.
[00:18:12] **Mike Bifulco:** it's a blessing and a curse I think.
[00:18:14] **Nikhil Gupta:** What's your favorite API you think out there?
[00:18:16] **Mike Bifulco:** So I'm, I'm definitely biased here because I also used to work at Stripe, so I was on the Derell team there.
[00:18:21] **Nikhil Gupta:** Yes. Fair.
[00:18:22] **Mike Bifulco:** obviously Stripe is the gold standard in many ways to work with. And I've, I've, you know, through the course of my career, especially working in developer advocacy for a while, played with many APIs, small and large to kind of build demos and show people how to do things and.
[00:18:36] I think a a So Stripes, API is wonderful. The documentation is incredible. The team is top notch, like super cool to see, but there are also smaller companies making things that are really interesting to use as well. Lately I've been looking at weather APIs for various purposes for my current company for craft
[00:18:51] **Nikhil Gupta:** Yeah, yeah, yeah.
[00:18:52] **Mike Bifulco:** the most thoughtful APIs that I encounter are ones where I don't have to become a meteorologist to understand how to use it. And that's, that should be the bar for everything. Like [00:19:00] my, my lowly self should be able to get to your API and understand what's going on, you know?
[00:19:04] **Nikhil Gupta:** Yes, yes. Yeah. Yeah. And like yeah, like that's. Also would love to hear from people. It's like, like I always wonder like when people show up, like like we have a swagger, right? Like they show up to this. Does everything just make sense? Like it just, if you
[00:19:17] just to look at the top level concepts that exist in the API, does everything come together? Yeah. Strip is great because you don't strive is very painful, but it rewards you like you go through the pain
[00:19:28] and because it's so complex. But everything kind of does make sense. I think that's like where. A lot of like teams like lose, I guess, like you, the more complexity you add can everything still seem coherent? Like as, as you learn? Does the world, does the world of your, does your world make more sense or does they make less sense because there's contradictions everywhere, like different ways to do things? Or is it just like there's one way to do thing and like, yes, it's very complex, but like if you learn everything, it'll make sense.
[00:19:59] **Mike Bifulco:** [00:20:00] Totally.
[00:20:00] **Nikhil Gupta:** does a great job. Great job of that for sure. I, I think there's something about like probably like, I don't know if like AWS counts, but I think a Ws
[00:20:11] is like is also kind of the same thing where, pretty complex, like very hard to work with. But like once you learn it, it's pretty okay.
[00:20:19] Everything makes sense. Actually, you know what my favorite API is Kubernetes.
[00:20:24] That's, that's Kubernetes? Yes. Kubernetes. I, I might even say I like it better than Stripe.
[00:20:30] **Mike Bifulco:** Oh, that's interesting. That's a new take for me, for sure. Yeah.
[00:20:34] **Nikhil Gupta:** It is, I find like the documentation when you read through it, like they, they read your mind. I think maybe that the problem is that Stripe has too many customers.
[00:20:44] **Mike Bifulco:** Hmm.
[00:20:44] **Nikhil Gupta:** So when, when you, when you enter Stripe documentation, like they're trying to walk you through how to handle like subscriptions, they're trying to walk you through how to handle the invoicing. They're trying to walk you through how to handle usage based billing. And they try their best and it's [00:21:00] great. Kubernetes, you enter the doc and they'd already know what you're trying to do. They know that you're
[00:21:05] trying to deploy something that can scale, so they talk you through different concepts. Like the first thing they'll talk about is here's pods, here's services. If that's, it's just the documentation gives me joy reading it,
[00:21:19] like I, it's a conversation with like, the person who wrote it is
[00:21:22] awesome.

## Docker vs Kubernetes: A Developer's Perspective
[00:21:23] **Mike Bifulco:** I, I wonder if you'll think of this as a hot take. But I, I see Kubernetes documentation and the, the way that they describe things in stark contrast to the way that Docker's developer experience is and was for a very long time. My, my feeling on Docker and figuring out how to make Docker work for, for a really, really long time was Docker can't tell you what Docker is, and you have to figure it out by trial and
[00:21:45] **Nikhil Gupta:** No. Yeah, I think Doctor has excellent like dev interface, like, you know.
[00:21:51] **Mike Bifulco:** Yeah. Yeah.
[00:21:52] **Nikhil Gupta:** Docker form makes it really easy to like, get things up and running. Like Kubernetes is harder to get up and running. But yeah, like, and [00:22:00] Docker logs for example, like everything sort of works, but yeah, like it's, they have no clue what they're doing.
[00:22:05] It's kind of the vibe that you get for sure.
[00:22:07] **Mike Bifulco:** Right. Yeah. You, you kind of have to know that Docker compose and docker's form exist to be able to figure out what they do. And it's not obvious what either does, you know, from the name alone. So let's talk about your, your use case.
[00:22:19] So you've, you've spent all this time building a, a great API. What is it like to use wapi? Like, what, what does hella world look like?
[00:22:25] **Nikhil Gupta:** Yeah. The hello world is and I'm curious to get your thoughts on it.

## Hello World with Wapi: A Hands-On Experience
[00:22:28] **Nikhil Gupta:** The hello world is you go to a dashboard, you don't even touch the API first, right? You go into the dashboard and the first thing you'll see is like and prompt to create an assistant. And you, you select like there's some templates options there.
[00:22:42] Like, Hey, I wanna make like a game NPC, or I wanna make like a customer service. I wanna make a, on my scheduling. You click one of those assistants, you get a template, and after that one click you should be able to, there's a button, right, that will show which like talk to assistant. And that is like [00:23:00] kinda the hella world there.
[00:23:00] It's like you have quit an assistant that you like. And you can talk to it. And the, what you'll see front, right front and the center is like, what's the configuration? Configuration of the assistant. It's like, oh, like my name is not Domino's. My name is like Pizza Hut. You know? It's like, welcome to Pizza Hut.
[00:23:18] Like that's changing
[00:23:19] that, that, that should give you the first like satisfaction of like putting, cutting the system and being able to talk to it. Then you, you know, the dev experience, which is like, okay, where do you wanna put this assistant? Do you wanna put it on a phone number? Do you wanna put it on your website?
[00:23:32] Do you wanna put it in your app?

## The Developer Experience: Making APIs Accessible
[00:23:34] **Nikhil Gupta:** I guess like that's where APIs and the experience, like dev experience enter. I haven't figured out the hello world for it, but that then you would go to our documentation and then they're kind of like. It's like, you know, choose your own journey.
[00:23:46] It's like I'm gonna create a phone number. Then you go to this doc, you wanna create like a put on a website, go to this doc.
[00:23:52] **Mike Bifulco:** I think it's an interesting case because there's a lot to be said for being able to just poke around with a thing and try it, especially with the case of what you're [00:24:00] building. Kubernetes is a great example. If I know that when I get the API calls correct for Kubernetes, that I'm going to have something that does all this nice orchestration for me, and the, the outcomes of that are clear from the onset.
[00:24:11] But for the product you're building, the outcome is like. A little more ambiguous and perhaps less clear in that, okay, if you execute these calls correctly, you now have a voice service. Is that any good? Right? Like the first thing you need to sell is, is the outcome of all this dev work going to be valuable?
[00:24:25] And I think inverting the experience in that sense is honestly probably a really good choice.
[00:24:29] **Nikhil Gupta:** That's like smart. Yeah. That's pretty smart observation. I've never thought about it that way. All I know is like as a developer, I think the thing that probably like now I know like I have to do Stripe, right? But like, I, I, wish Stripe just gave me like more joy, like before I even get started, like, do I, will
[00:24:46] stripe even work for me? I think it takes me a long time to figure that out, that answer to that question for most dev tooling. Whereas I think we are trying to answer the question like, yeah. Bobby, will this work for you? This is, here's a assistant, and then it's your, the, the job is to [00:25:00] just figure out how to like, get that assistant in the right place that you need, which we have all the AP ideas for.
[00:25:05] But that I, I think about the law too where like, it's like, what is like in my, like where dreams, like what would I love as a developer from this experience? You know, it's like, what is like the, the, the most beautiful, amazing experience coming into Wapi? I don't have an answer to that question, but like, yeah.
[00:25:21] The hello world is don't code. Just go in,
[00:25:24] try something and then if it looks good, get, get, get cracking.
[00:25:29] **Mike Bifulco:** I think it's pretty compelling, and I, especially for this audience that we're talking to. So for API developers, you get two things out of the box before writing your first line of code. One is you can go and try the thing, right? You, you go to the dashboard, do that, do your own experience. But naturally, I think most of the people listening to the show will also go to your API.
[00:25:47] Reference, right? Go to the docs themselves and see what's there. And like, there's a lot of good, juicy stuff in there that's, that in itself is exciting. And you can kind of see some of the thought and, and usability that comes outta that straight away. And to your credit too, there's a handy [00:26:00] little button on a lot of pages on your site, maybe all of them that says ask ai where you can maybe get some answers, some questions answered pretty directly too, right?
[00:26:07] **Nikhil Gupta:** Yeah, that's awesome.

## The Future of Dev Tooling with AI
[00:26:09] **Nikhil Gupta:** Like too, where I think now this like dev tooling, like, like there's, so, I think there's gonna be so much, like, so much rich experience around like dev tooling because a lot of it more, a lot more can exist thanks to like ai. Like,
[00:26:23] like before, I think like if I was to think about like integrating a new API, it'd be like, oh man, I don't wanna figure out the documentation.
[00:26:30] It just seems like insane. Like, that sounds like a lot of work. Now I can go to the documentation, talk to the AI and be like, okay, like I need to do this. Can you gimme the code for it? It gives me the code I just put in my code. Like that
[00:26:40] is transformative.
[00:26:42] I think to like composing, like composable APIs, I think we're gonna see a. So many APIs, I think in this like next, like two, three years it's pretty, pretty exciting for me. But yeah, I think for people like, who are considering, I guess like yeah, like what would [00:27:00] for API builders, I guess, like the question you, the reason, the way you would, the reason you. It's like, maybe you should ask yourself if, like, what are you trying to build?
[00:27:09] Like, are you, A lot of our users are startups, right? Because they're trying to explore prototype, whether they wanna sell into like a small, like a particular niche, like Vertical sa or I guess like the, like everyone knows Activity four Oh will be everywhere. Like that experience that this showed, I think that's going to be in every phone. Every microwave, every car those
[00:27:33] assistance. So now the question is do you wanna bring that experience? Do you wanna help get that experience deployed in different places and you can wait for four to come out, or you can use to get started on that right now. And then when we do have gbd four, we'll just plug it in and then you'll just have, won't have to do any work to switch it over. And, and then I guess it is for us to figure out like what other value we can add, [00:28:00] which will be like observability, tooling. Like you feel like a million calls that you're doing, are you gonna really use like the raw for opening API or are you gonna use like. Some production ready tooling around to make sure that your calls are going well.
[00:28:16] There's a visibility, tooling, but that kind of thing is I think people, what people would use this for and that's what we are here to provide. So yeah, I think I'm drawn to people like if you think voice cool. If you think multi model is cool, give it a try. Like make something prototype. It's so easy to prototype nowadays.
[00:28:33] **Mike Bifulco:** Sure, yeah. The imagination should run wild and, and go build the things that you definitely wouldn't have been able to, you know, in 2022 or, or however you wanna look at that.

## Use Cases and Real-World Applications
[00:28:43] **Mike Bifulco:** Can you tell me about maybe an interesting use case or two that you've seen? With api, I.
[00:28:47] **Nikhil Gupta:** sure. I can tell you the UCSA. Yeah. WC this is our first customer that we worked with. They do sales training. They're called hyper bound. And you can go to their website. It's really fun. It's [00:29:00] called hyper bound. I don't know the actual domain, but so hybrid, it will show, it should show up and you can call these boss to practice selling. So the founder, you can imagine, like, it's pretty learning how to sell is like a pretty key skill and pretty hard. You can practice on real humans, but then you're leasing losing real deals. So with practicing these bots, and these are bots are really hard to sell to. Like they would like hang up you on you all the time. So that, that's an interesting kind of role play is a very interesting use case that we see where like having AI to represent like hard conversations, having represent like sales training. So that's like one category. The other one that's pretty obvious is customer service, where people call into this, this restaurant, it's like, I want pizza here, or I wanna book a reservation.
[00:29:46] Like, here's an availability for the reservation. That kind of like function calling, tool, calling use cases are another, what are other popular ones?
[00:29:55] **Mike Bifulco:** I am. I'm curious you mentioned your founder created a coach bot early on. [00:30:00] Do, do you or your co-founder have any that you use for yourselves? Maybe internally, externally, personally, I.
[00:30:04] **Nikhil Gupta:** I, I use for testing my, my Kill bot.
[00:30:07] But I don't know if I do, I have like I mean, there's one thing which is I don't like the experience of like CGBT, like voice.
[00:30:17] As much, 'cause I mean, right, right now they haven't deployed it before, I think in production for voice and being able to kind of brainstorm because like I'll get stuck on a problem and I just need to
[00:30:31] like, talk to something. It's like. I'm, I need to think out loud with someone that's like intelligent and used, I used to do this with my co-founder all the time, and now I need to do
[00:30:40] less with him, just with the ai. I dunno if that, what that speaks for human connection, but it's just like, yeah, just, I just loop it out and then I talk to it, but like I'm just holding it down because I'm still processing and I just like pick up a thing and it response. So brainstorming is a very common use case for me.
[00:30:56] And yeah.
[00:30:58] **Mike Bifulco:** have a I have a weekly [00:31:00] newsletter that I write on my, my personal site under my own name, Mike, by full I publish a newsletter for startup founders and JavaScript developers. And often I'll sit down in the morning and just brainstorm, like talk, paddle through some ideas and try and you know, expand the idea to something that kind of has a beginning, a middle and end.
[00:31:16] And yeah, I, I agree with you. I don't think 4.0 is deployed yet to that audio experience, and it's a little irritating when it doesn't work well. The, the biggest change and the most valuable thing for that is that now it has memory context where I had previously been like, fine tuning things over and over or re-uploading transcripts.
[00:31:31] And now, now at least that part is getting better and that's been a big helpful change.
[00:31:35] **Nikhil Gupta:** That's awesome. Yeah. Yeah, exactly.
[00:31:36] I, I think my co-founder and I have a bet around like whether as like years go by or months go by, people like. Will there be a wide widespread love for AI or widespread like skepticism around ai? My, like, it's open, open question because we will be like, oh, AI's getting better, taking my job, or a lot of negativity or responsible, like, I just think like when you look at activity four, [00:32:00] oh, like talking of the thing is like magical like that demo and it'll be hard not to like that experience and like want that thing.
[00:32:09] Everywhere. So very curious to see that, that sentiment shift on ai what will happen. But thanks. Four.
[00:32:18] **Mike Bifulco:** It is hard to tell, and I think that the mindset is very split right now in many directions, old and young. I think men and women have different feelings on this. I think different generationally, you know, gen Z and millennial and, and all that. There's, there's lots of different opinions on this stuff, and I think
[00:32:32] we'll figure it out over time.
[00:32:33] Yeah.
[00:32:33] **Nikhil Gupta:** Yeah, but I guess like, I don't know if you run into this, but like, for example, my mom, she's like not very technical savvy. And for like, for example, like she needs to, wants to learn how to, like, she's figure this out now, which is great. Like how to post something on Instagram. And the process for that is like you sharing your screen and you're just kind of pointing her to a button, like press this. Talking about it, like, to me it just [00:33:00] seems like, like yeah. Would be so good at that. Like, I think all people would love it. Like, it's like it, it's patiently just kind of looking at her screen and like talking you through like how to like, do something
[00:33:11] **Mike Bifulco:** Yeah.
[00:33:11] **Nikhil Gupta:** that is pretty compelling.
[00:33:14] **Mike Bifulco:** Yeah, a very a big potential for a tender and loving experience that helps you grow at your speed which is super interesting. Yeah. Yeah. So cool.

## Technical Deep Dive: Building with Node and Kubernetes
[00:33:23] **Mike Bifulco:** Tell me a little bit more about how you're building Matthew.
[00:33:25] Like what, what is underlying your your software?
[00:33:28] **Nikhil Gupta:** So, yeah, we decided to use Node because the thing, so what's that? Like, why is voice harder than chat? Right. And like open A is now running into this, like getting, trying to deploy multimodal systems is because the chat it's synchronous like, it's like one thing after the other thing that after the other, the voice the person is talking, they might stopping stop again at any point and then you start talking, but they might interrupt again.
[00:33:53] So there's like a very dynamic environment which makes it hard to write like code for it. [00:34:00] And so for us, like given that there's so much like, like there's, you're getting audio every 20 milliseconds. You given that there's so much activity, you clearly need the event loop, right? You need like event loop, like audience coming in, like process that is there something you wanna say?
[00:34:15] Like send that out? If you need an event loop now for an event loop, you could try to build, use Python, you could use something else. You can make custom event loop in rust. But, but I was like, no, no is such an obvious answer for me because. I'm standing on the shoulders of Giant, like this event Loop has been optimized the heck out of it.
[00:34:38] Like, because the entire world browsers running this thing. But the event loop is really good, basically like that. I cannot, I can't imagine all to do anything better. So we use Node to manage the asynchronous nature of our workload, which is like you're getting on audio base all the time. You, you're sending out audio all the time, and there's like so many checks between those two. [00:35:00] So the a harass nature node, these stack for the API interface, we use nest js.
[00:35:06] Because Nest has like an excellent experience around like structuring, like making sure they get very opinionated interface around like how to get data moving and get, get, get operational. And then also exposing that, like, exposing that in docs, exposing that in like exposing, having validators on it. So Nest js is like kind of our choice for an opinion and framework in Node, like we can run in node. And then for the database side of things, like I love Postgres. I've always loved Postgres. It's just awesome. Never had any issues with it. And then so we use base for it, I think has an awesome managed offering that. Kind of, I think, gets to the heart of it where you have nest node. And then of course we have like, kind of a we have many different like GPU services that need to run. Like there's
[00:35:55] things that you can often turn on, like, for example, you wanna turn on emotion detection. That's a good GPU [00:36:00] service. You wanna turn on Denoising, that's some other G service. So we, our cluster is like this main API gateway. Sharding to like different workers and different services for different stuff. And we use Kubernetes so as everything because That's great. Oh, one thing I think that's kind of underrated that we love is Lummi.
[00:36:23] So if you use, you've probably used Terraform before to spin up everything. Lets you do that in your programming language of choice. And we, I love it where it's like, you can imagine like, so for example, we have many different reds that we use. We specify, we specify a list of them and then we can just literally write that code. Or cons, Redis of red. Like, it's like, before I do this, like it's like, and it'll actually do it. It will, it just works. It's crazy to me, like that, that is possible. So Lummi for orchestrating things [00:37:00] yeah.
[00:37:00] **Mike Bifulco:** That's an unsung hero. And so I, I think you mentioned before you publish an open API spec with swagger. Is that right too?
[00:37:06] **Nikhil Gupta:** Yes, yes.
[00:37:07] Do you want me to link to it?
[00:37:09] **Mike Bifulco:** Sure. Yeah. I'll make sure it's in the show notes. You,
[00:37:11] **Nikhil Gupta:** Yeah. Yeah,
[00:37:11] **Mike Bifulco:** welcome to drop it in here.
[00:37:13] **Nikhil Gupta:** Yeah.
[00:37:14] **Mike Bifulco:** That's something that our audience will be super interested in. It's we talk a lot about open API in that world here for
[00:37:20] **Nikhil Gupta:** I love open a PII like, I think this is the concept of like, like I just envision, you know, everything to live in my code. I editor, like, I just create this thing and then this open API like is like kind blossoming into like this types everywhere. Kind of the, the dream world for me it doesn't exist right now as much.
[00:37:39] I don't think there's like as yeah, it, it getting there, but open a open API is awesome.
[00:37:45] **Mike Bifulco:** Yeah, it's, it is super helpful. We we have lots of hard opinions on, on when and how and why to use it around here. And this is one of the things I like about this community is I'm always learning from people smarter than me about, you know, tools and things available in that world too. Cool. So, Nikhil I'm, I'm curious what's next for you?

## Team and Future Plans
[00:37:59] **Nikhil Gupta:** [00:38:00] Yeah. For us, I think more observability, more monitoring kind of like going into Wapi and then being able to see how your calls are doing knowing that they're doing well. That is kind of like what's next for us, like giving people security that like, like, their voice, AI is doing the work for them.
[00:38:20] I think that that would be a big thing going forward too, where like, you have all these agents, but like how can you be sure that they're actually doing what you want them to do?
[00:38:27] **Mike Bifulco:** Yeah, making the black box a little less black is a very, very valuable feature especially when your business relies on it. If I'm standing up a customer service endpoint of some sort, I want to know that it's doing great customer service.
[00:38:38] **Nikhil Gupta:** So for us it's
[00:38:39] like all about like that, that satisfaction of knowing it's, it's working well.
[00:38:44] **Mike Bifulco:** Super, super cool. so we've talked about how you build things. We've talked about yourself and your co-founder. How big is the team right now?
[00:38:50] **Nikhil Gupta:** We are a team of like six right now.
[00:38:52] And definitely always looking for the next great hire. If you love making, if you love [00:39:00] voice ai and are curious about like this whole world of modality and getting the deployed. Which we feel really passionate about and like making great experiences for developers to make it easy. We get it deployed everywhere. Yeah. I would love to chat.
[00:39:13] **Mike Bifulco:** Yeah. Perfect. That's great.

## Conclusion and Final Thoughts
[00:39:15] **Mike Bifulco:** And so just, just to wrap things up then, where can people go to find.
[00:39:19] **Nikhil Gupta:** Ai, we, we are kinda everywhere where you can find us on Twitter, LinkedIn, ai. We have a blog. Again, if you're a developer and you love docs, you know, this is actually, this was a learning for me. People don't go to our website. Like they'll just go straight to go to our docs. Like, they don't go to the dashboard, they don't go to anything.
[00:39:35] They, there's like, oh yeah, just like read through the docs. I'm like, oh, cool. That's that's, that's interesting. So docs, we have ai, just Wapi V be
[00:39:46] the.
[00:39:46] **Mike Bifulco:** have definitely found yourself in the right place to get to people who want to go straight to the docs and read. Of course, will make sure that there's links to everything you just mentioned in the show notes.
[00:39:55] **Nikhil Gupta:** Thanks.
[00:39:55] **Mike Bifulco:** Nikhil. Thanks so much for joining today. It's been a real pleasure chatting with you, and I'm, I'm really excited [00:40:00] to get in and poke around with some voice assistance myself.
[00:40:02] Please, please feel free to join us anytime if you have more news to share or things that you're interested in chatting through and sharing with the audience. We'd love to have you back. Thanks a ton for joining today. I appreciate it.
[00:40:12] **Nikhil Gupta:** Having, ​