- Duration: 32 min
- Platform: zoom
- Source: Read.ai 01JBFAYBF13BFK3Z2X7PH0K8P6 (https://app.read.ai/analytics/meetings/01JBFAYBF13BFK3Z2X7PH0K8P6)
Transcript
[UNKNOWN_SPEAKER]: By the way, feel free to kick this thing [Sahil Manocha]: I mind, the note taker. [Michael Houck]: Yeah, sure. Totally fine. I don’t really care either way. [Sahil Manocha]: Okay, cool. You have been been a minute. How you been? [Michael Houck]: Yeah, man. Been good. It’s crazy reconnect. But sounds like you guys are doing some cool stuff. I checked out the site looks super interesting. What’s the journey been like since hack house? [Sahil Manocha]: Yeah, now it’s been it’s been long. But I will tell you one thing across the office. I have Kaelin from hack house. I have so harsh from the same hack house. No way. I met Julia like seven months ago. She started a company as well. I know Kian from Hack House started a company. Poojan is across the hall as well from Hack House. [Michael Houck]: I saw Poojan in SF actually like a year ago. And we kept in touch a bit. Might have him on our new podcast. [Sahil Manocha]: Which one? Who? Who are you talking about? Poojan. Oh, Poojan. Yeah, yeah, yeah. Yeah, no, it’s a lot of people from that, from at least our cohort. I’m not really sure of the others, but yeah, a lot of people started companies and they’re doing well. [Michael Houck]: Yeah, I mean, that’s what we wanted. [Sahil Manocha]: So it’s amazing. I don’t know what the stats on LaunchHouse are because HackHouse stats are, I mean, clear as day for me at least. What does it look like for the LaunchHouse cohorts? [Michael Houck]: Yeah, I mean, overall, everyone who was doing LaunchHouse was like already building a company, had an idea basically. Them in, um, or they had like quit their job and knew they were going to build something. So in terms of like stats, you know, we’ve had people at this point, like well into double digits, like 20, 30, 40 plus people going to raise from like tier one VCs. And we wrote like 50 checks out of the funds as well ourselves. Um, but I left the whole thing in like Q1 of 2023, I guess it would be so. I’m not involved in all that anymore. Um, I think Brett and Jacob still run the fund, but they both have their own companies too. So I don’t think it’s like a main thing for them either. [Sahil Manocha]: Yeah. I’ve been seeing Brett’s posts. So I think he’s, uh, he’s hard at work. [Michael Houck]: Uh, yeah. The AI, uh, AI email thing is pretty cool. [Sahil Manocha]: Yeah. Yeah. Cool. Awesome. Yeah. Let me, uh, jump right into it. So yeah, we, uh, after hack house, I, I was still at Google and the year after I, moved, like late last year, I moved to SF to start a company. So I did this thing called OnDeck last year. Oh, I know OnDeck. I met my co-founder during OnDeck and he was in New York while I was at Google. We hung out a bunch and late last year is when I moved to start, to start ArcSpeed. So we didn’t start until, in earnest, until April this year. So we were exploring, I quit my job early March and that’s when, that’s when things, you know, started. So like six months so far, basically. [Michael Houck]: Yeah. Yeah. Pretty much. Cool. [Sahil Manocha]: And we launched like, not really a launch, but we did like, we started taking customers in a month ago and yeah, just been working, you know, fixing things since then. Still, still not scaling yet, but it’s going to get there. What I can do is I was first off, like it was so serendipitous because your email was surrounding you know, the voice AI thing. And I was like, Oh yeah. [Michael Houck]: When I, when I saw that’s what you guys are working on, I was like, Oh, we definitely should hop on a call and chat like ASAP. [Sahil Manocha]: Yeah. And we’re in the middle of our, of our seed race. So, you know, I think timing wise, this is perfect. Let me, what I’ll do is I’ll, I’ll pull up the deck and I’ll walk you through it. Of course, we’ll talk about history of, of me a little more, my co-founder. And we can dive into why we started the company, what the motivation is. Where’s the round at? [Michael Houck]: Like you say, you’re in the middle of it, like how much is committed, like before we dive in, just give me the kind of the stats. [Sahil Manocha]: Yeah, so we’re doing three and a half on 20 cap. This is our seed. So we did a pre-seed from Pear Ventures in SF. And now for the seed, we have about 700K committed from Angels. Angels that we like, but we can reduce the allocation based on a lead, based on the allocations that leads usually want, one targets and so forth. But yeah, right now we’re in final conversations with a couple of funds and yeah, they haven’t closed yet. So looking for a couple of leads at most two, and then we’re open to taking smaller checks as well, but that based on strategic alignment and things of that nature. [Michael Houck]: Cool. Totally. Um, to give you some context of sort of what I’m doing with this right now. So, uh, you know, Ryan Hoover, right. Founder of product hunt. Um, yeah, so he’s a friend and we’ve been working on some deals for weekend fund, which is his fund. Um, and so this is related to that. So we’d be writing checks out of weekend fund into, uh, into you guys. Um, so sector agnostic seed, pre-seed, both are fine. So obviously in this case, you guys are doing the seed. Check size would be like 200 to 400. So probably with the size of the round you guys are doing, probably wouldn’t be a lead, would be a smaller check, but high conviction team and Ryan’s a high conviction guy. So if he’s excited about it, he’ll pull a trigger regardless. [Sahil Manocha]: Okay. Sounds good. Yeah. Let’s, let’s walk through it and we’ll come out the other side and discuss specifics afterwards. Sounds good. Cool. So we are Outspeed. We help developers build real-time apps on voice and video. And just to walk through the backgrounds of the team, you know, I was born and brought up in India, went to Launch House, the biggest highlight of my life. Outside of that, you know, I went to the school in India called IIT and then Columbia University here, did CS in both places, did some deep learning research. And after that, joined Microsoft. I was building network infrastructure there and then joined Google. I was optimizing compute workloads and quit my job around February this year, around March this year, met my co-founder a year and a half ago. Just to dive into his background real quick. He was born in India as well. Went to IIT briefly. He was actually a bit of a celebrity around college time. He ranked second in the whole country in the IIT exam. Yeah, so if you have any Indian colleagues, they’ll get a kick out of that. No, I know. [Michael Houck]: I know how big a deal that is. [Sahil Manocha]: He was on, he was on, on newspapers and, uh, and in TV shows asking him, you know, how did he do this great feat? Uh, I ended up going to MIT though. So he was at IIT only for a year, ended up going to MIT after, uh, did CS, bachelor’s and master’s there, uh, did some self-driving car research. And after that, I joined this company Autogrid, uh, which was 15 people at the time that he joined and 150 people when he left. So we really saw the growth of the company from small to large and the company got sold to Schneider last year and he also quit his job around March this year. When we met, you know, we weren’t looking to start a company together or anything like that. We became friends first when he was in New York and then late last year when I moved, I was in the same headspace as him and we were looking to start something at that point. So we met for coffee, started working together and that’s how Oxpeed, you know, that’s where the story began. So one of the first things we were doing was building customer support voicebots for e-commerce companies. So we’ve got people in our network who run e-commerce stores selling technically complex products like home security systems, camera systems, routers. And unlike regular e-commerce, they get a lot of call volume because they’re getting calls from small businesses, schools, organizations asking about integrations, tech support, installation help. And so they had to deal with these calls. So we were automating that for them. And we realized two things in doing so. One is we’re not the best founders for e-commerce. It was just like, bad farm market fit for that market. And second, there’s a lot of foundational problems in voice infrastructure that had to be solved to build that application, right? And not just for voice, but for video as well, because we could see multimodal models on the horizon. So if voice understanding has gone up, and that’s what voice applications are built on, video understanding by models has also gone up, right? And so now you can start building real-time applications on video as well. Now, also, by the way, I mentioned self-driving cars before because they are the ultimate real-time application in the wild right now. They ingest huge amounts of voice video sensor data and react with very low latencies. So people don’t make the connection between voice and self-driving cars in that way, but the network primitives and the compute primitives are very similar. So just to talk about what the market map looks like, of course, on the left in these four boxes, you’ve got what are conversational AI companies, right? And these are essentially operating off of conversations in some shape or form. Of course, some of them don’t only have real-time use cases. They have non-real-time ones as well. Builds like LND videos and translation and such, but they’re getting into real-time as well, right? And any companies that you’ve spoken to in this map? [Michael Houck]: In terms of use the product, I’ve used all the note takers except for Granola, right? And, and Naka, I don’t know Naka. I know of, and have seen used like Gong, I’ve used Hagen, and I think that’s it. Okay, solid. Cool. Yeah. [Sahil Manocha]: So the goal of this slide is just to show, you know, these things came about two, three years ago at most. And they’ve exploded in popularity, right? And there’s a lot, and I’m only talking about like the real-time use cases. So Apriora, for instance, is an AI interviewer, right? So companies like Mercore and MicroOne are doing that to scale their operations. On the right, these boxes are, you know, arguably bigger going into the future, and they’re based off of the same primitives. You’ve got really fast networking, streaming infrastructure, and then intelligence as close to the application as possible, right? So the models and the orchestration layer runs close to the consumer of these apps. So Wearable AI example, which is now infamous as Humane, for some reason their latency was dog shit, pardon my French, but it worked on the same principles. Now, if you look at the building blocks of a real-time applications, there’s really three things that you need to solve for. And ask me if you need me to, and let me know if you need me to zoom out more or dive in more based on the technical details. Right? But the three buckets are really the models, right? So this is where open AIs, real-time APIs, speech-to-speech models, or 11 labs, or HeyGen, all of those guys sit here. Then on top of that is what is called a framework, right? So this is where developers are bringing together different things like RAG, integrations with databases, auth systems, and open source models to power the business use cases for the application. This is also called a compound AI systems. If you ask a VC, you know, or like even in the developer community is getting that name. The third piece is the infrastructure. Once you’ve written this application, you’ve made sure every piece is close together to reduce the latency, then you want to place this close to the user. If you’re in New York, I’m in LA, and someone else is in the Middle East, the application must run in those areas to service the low latency use cases. The second thing that is underappreciated but needs to be built from the ground up is monitoring for these apps. These are operating on voice and video inputs, right? They’re not operating on text inputs. So if something goes wrong, how do you know, how do you replay it? How do you test it? Well, you’ve got to figure out what the input was and what the error is. And the input, if it was voice or video, you can’t see it on a log line, you know, in Datadog. So these tools need to be built from the ground up. And third is like the networking for these applications is very different from how the web is networked right now. So they use WebRTC, so really fast types for voice and video. And the SDKs to embed these applications on the front end are also different. Right now, by the way, when we’re on a video call, we’re using WebRTC, right? So video call infrastructure. Sorry? You’re using what, sorry? We’re using a real-time protocol, really. So on the web, it’s called WebRTC. On the cameras, it’s called RDSP. There’s a bunch of different protocols. But the point is, they’re hard to work with, and they require specialized infrastructure, right? That’s why there are companies doing it specifically. If you look at the market now, if you look at what developers are doing right now, they’re Frankensteining these things together to build real-time applications. So they take providers like LiveKit, Daily, Agora, which are WebRTC providers. By the way, OpenAI uses LiveKit and Agora, right? Retail uses LiveKit, Vapi uses Daily, Tavis uses Daily. So these are essentially fast networking network providers, which move voice and video bits from the user to their backends. Now, if you’re an application developer, you can’t For you to build an application, you have to integrate with these guys and run your backend and scale your own backend yourself. So you will run it on an IaaS provider like CloudFlare or on AWS or Azure, whatever. So you’ll run the backend yourself and figure out a way to stitch these things together. And the second thing is you’ll have to set up monitoring too. And you’ll bring these things in a way that is low latency. Some of these are code level things and some of these are services. So say you have a vector database and you’re running application in India, the vector database must also run in India. But you’re a developer that’s servicing these applications across the world and you don’t want to be deploying things in India and then Bangladesh and then U.S. And different parts of U.S., right? So there’s orchestration that is required. That’s where RxSuite comes in. We’re a purpose-built platform as a service for real-time applications on voice and video, right? The way it looks for the developer is they write the application using RSVK and then they deploy it from the single line command. And this is really the soft, this is the money right here. It sets up the monitoring, the infrastructure for networking, the computer infrastructure. It scales based on number of conversations or interactions that your application is having concurrently. So you don’t have to do any of the management yourself. Any questions at this point? Nope. Yeah. By the way, if you’ll use any of the voice providers, right? So there’s like Vapi, Retail, Volcode is Kian’s company. I think Kian’s pivoting now. Because of some execution things. But Bland, Vapi, and VTL are doing the same thing. Bland is more application layer, so they’re doing no-code tools. They’re not targeted towards developers. Vapi and VTL are. The problem with them is they still require you to run your own backend, right? And it doesn’t really take away the ops pain for developers. And the other thing is, since they rely on LiveKit and Vapi relies on Daily, by the way, those companies that are eating into their business. So, whoops, LifeKit and Daily are going down the stack, right? And what ends up happening is when a customer’s needs become more sophisticated, they get off of retail and Vapi, they build out their own stack, and then they integrate with LifeKit and Daily. The problem is that these companies are operating in the framework layer. And this layer is the most fast moving thing, but it pretty much depends on the developer, where the developer world is going, right? We don’t, we, although we an SDK here, we’re not dependent on that, right? So that might change and we’re not, we’re agnostic to what SDK you use. Anyway, so we launched about a month ago. All the people on the right are our customers who’ve paid a platform fee of $250 to use the platform. Now that will go away. You know, it’s, it’s not typical for a pass to charge a platform fee, but we’re doing so to screen for ability to pay. [Michael Houck]: Yeah. Smart. Unreasonable pricing up front. [Sahil Manocha]: And it is unreasonable. I will admit, and it is causing us to take on fewer customers and, you know, but those things are required because the infrastructure needs to be built out and made robust and it’s a little wonky right now. But yeah, so DeepReel is a HN competitive that uses us to build real-time avatars. You’ve got Echo, which is a productivity tool, watches your screen and suggests insights. GigaML is a customer support company. Sylvie is a language learning company. We have another agent’s company and one avatar company that onboarded recently. Avatek also onboarded recently. So ZoomInfo in particular is a bigger get, right? But they want us for an AI sales agent for top of the funnel sales. And we just got done with SOC2, so submitted all the security stuff with them. We’ve been engaging with them for about two months and hopefully things will convert. It’s mostly based on pen testing. There’s a lot of specifics that need to be figured out. But there’s heavy investment on their part in terms of bringing their team in. And we’ve had multiple meetings with them already. So we launched On Product Hunt. So we did that about a month ago. Got product of the day number three, which is high for an infrastructure company. Got about 700 signups off of that. My co-founders working through them, qualifying them for ability to pay and for the use case. We also have a community of 400 people 150 people on Discord. We actually did an event with OpenAI, Cartesia, and Zoom as partners two weeks ago at AGI House, got about 200 attendees off of that. So that’s the bottom-up sort of PLG motion that we are starting that flywheel going. In terms of go-to-market, for the initial phase, it is PLG. The second is like partner sales. So DeepReal is a model provider. We’ve also partnered with Viviate, actually their CEO is an investor in us. And so these will serve as essentially channel sales partners. So anybody who wants to use their models or their infrastructure with real-time AIs will host on us. So that would be in the way of content and in the way of recipes on their platforms. Right. And the third go-to-market strategy is going to be enterprise motion. We don’t have a repeatable playbook right now, but GoZoomInfo actually came in via PLG. Right. So traditionally, you know, with these companies, their developers kind of look at what’s out there and then move it up the chain. [Michael Houck]: Do you think it makes sense, like how much time are you planning to spend on the enterprise side, like Zoom InfoSight’s clients versus smaller, newer companies? So how much time are we planning to spend? Yeah, like how big of an emphasis is it for you guys in terms of how much time you’re spending and also in terms of your projections and the goals you want to hit? [Sahil Manocha]: I mean, one thing is like for the first six months, PLG makes the most sense because you want developer mindshare, right? And even in the enterprise case, if you look at comps in different industries, so Purcell, for instance, or if you’re looking at, you know, other past companies or developer tools, you need to win, even for enterprises, you need to win the developer mindshare for their developers to kind of move it up the chain. And a lot of those decisions are made bottoms up, not top down, right? So yeah, for the next 6 to 12 months, we’re focused on PLG while also onboarding enterprises. We don’t want to give up on that. So essentially our tooling will focus on enterprises, but our content will focus on PLG, right? With changes like the engineering resources going one way, but the DevRel and all the marketing goals for the PLG motion, right? Yeah. I think that’s smart. [Michael Houck]: You don’t want to be beholden to these bigger companies where deals can fall apart for things totally outside your control. [Sahil Manocha]: Yeah. But I do admit they will be the sharks, which will essentially, I mean, if you crack that, you’re in a better position to weather any sort of big player entering the market and then upending the PLG motion. Now, the way we price is very simple. It’s a cent per minute for usage of the cloud and four cents per minute if you do use the cloud and host the models on our infrastructure. And by the way, we are not an inference provider, right? We are only hosting the models for use within the applications that you deploy in RF infrastructure, which changes how we’re viewed. For the application developer, the benefit is they get better coverage of monitoring because we’re covering more part of their stack. Second is they get the best latencies possible because of this, because wherever the application is running is where the models must also run for the lowest hops. Third is bundled pricing. And really the tailwind here is if you believe that open source models will get good enough to rival closed source models for voice, for transcription, for video understanding, which has proven to be the case in image generation and has proven to be the case in transcription, there’s no wonder it would happen in others as well. Our focus is just adopting what’s the best out there and optimizing that for latency. We’re never going to train our own models. It doesn’t make sense for us to do so. So we’ll have them available in the model preview. The ground, developers can bring them in and use and deploy them. Now, if you look at the competitive landscape, we’re directly positioned against these companies, Livekin and Daily. These are WebRTC infrastructure providers. They came on the backs of essentially the promise that every company will have their own Zoom. So during COVID is when they had their peak, so to speak. And that’s when they raised the money there. The infrastructure is built around that as well for people to build their own meeting infrastructure. And that affects the developer experience that they’re able to offer, right? So companies find or users of their product find their developer experience to be subpar because their abstractions are all around meetings and participants and meeting rooms. But that is the only solution that exists in the market. So clearly people still gravitated towards them. And the second thing is they don’t provide the deployment piece. So although you can use them for the networking, their code must still run on your infrastructure that you have to manage. So the infra operations pain is not alleviated for developers and companies, which is like the maintenance cost is what makes things like ours compelling, right? And monitoring is unsolved. [Michael Houck]: Do you think like if other people come into this market who are more equipped for the developer experience and all that, do you think this is going to become a price war or do you think that there’s something that you guys have that would kind of be defensible in that way? [Sahil Manocha]: I mean, platforms don’t necessarily engage in price force because you’re offering bundled things. And so it’s hard to put a price on like the individual pieces and they don’t take out the pricing in that way as well. In terms of the other players that we see entering, I think Cloudflare can enter eventually, but they have such a wide IaaS offering that it doesn’t make sense for them to verticalize. If anything, I think these are best positioned to compete with us, these companies, and they’ll eventually move in the same direction. And they are moving in the same direction. The only thing is they are having to develop an orthogonal stack. So one for meetings and one for AI real-time applications. So they’ll have to cut off one of them at some point. But just to add this particular point, they are positioning themselves for real-time AI. Think. So I’m guessing like they on their roadmap, they’ll probably cut off the WebRTC piece at some point, or the meeting infrastructure at some point. But majority of the revenue comes from that. So yeah, it’s like whatever their priorities are. In terms of price force, I think monitoring is where you won’t have to be beholden to any sort of price war. Because you can charge per seat pricing for it, you essentially capture the data that users have on your platform at this start at your platform. So 0 to 1, 1 to 10, and 10 to 100, if they’re on your thing, your offering becomes very sticky for them. Because essentially, you put all their data that has flown for all the errors in the past and all the user data on your platform, which is something which is hard for them to move off of But maybe if there’s anything I can clarify on that front, maybe I didn’t. No, the answer makes sense. [Michael Houck]: The answer makes sense. Yeah. [Sahil Manocha]: Also, it’s like for a managed service provider, at least the way I’ve seen economics play out is if you’re just a thin layer, even though it’s a performance layer on top of something that the hyperscales offer, there’s not a lot of strength. You’re not operating from a position of strength. And that’s, by the way, that’s why a lot of the WebRTC providers, although they have a lot of people using them, They’re not able to charge. They don’t have great margins because they’re just charging on top of egress cost by AWS. So they measure their cost in gigabytes of data transfer, and so does AWS. So it becomes a race to the bottom. And so some company comes out in India or with cheaper developers and with enough time is able to catch up and is able to provide cheaper pricing. With platforms, that doesn’t really hold true because you have to not only solve for the infrastructure, which increasingly becomes a less important piece for bigger companies. For them, it’s the developer experience, the integrations, the workflows, and the monitoring that you offer, right? [Michael Houck]: Yep. [UNKNOWN_SPEAKER]: Yep. [Sahil Manocha]: A hundred percent. Anyway, that’s Outspeed and yeah, that’s the end of my presentation. [Michael Houck]: Oh, cool. All right, sweet. Look, I think it’s really compelling. I think it’s interesting. I love where you guys are at in the market relative to what’s out there. I guess the one last question for me is just, what’s going to stop you guys from scaling super quickly? What’s the bottleneck? Where do you see the issues coming up, the risks, all that? [Sahil Manocha]: So we have to really nail the developer experience. What will stop us from growing quickly? I think if first of there’s market risk, you have to believe that multimodal applications or at least voice applications would be The second is if they’re not sophisticated, if they’re simpler use cases only, it’s going to be limited to appointment scheduling and answering a restaurant canceling or whatever. Then that is also a risk because then providers which are closer to the application layer are more competent at that. They can provide a low-code experience. I don’t believe that to be the case. I think these applications will be more Um, I think you’re right. I think you’re right. So those are the two risks. Like if the applications remain very simple, then we are an overkill, right? And your problem. [UNKNOWN_SPEAKER]: Okay. [Michael Houck]: Um, look, I’m interested. Um, I’m going to chat with Ryan, see what he thinks too. He might also want to chat. Um, so I guess, can you send me over anything that you’d like me to include in my note to him? I’ll shoot him a note later today. Um, and we can go from there. [Sahil Manocha]: Yeah, I’ll send over the deck. If you want insight into the margins, I can send that over as well. We’ve got like 80% margins on average across the board. Well, yeah, the more the more info, the better for sure. Cool. Yeah, I’ll send the deck something on the margins and happy to talk. [Michael Houck]: Cool, dude. Good to reconnect. Excited for you guys. And yeah, we’ll be in touch soon. [Sahil Manocha]: Yeah, man. Likewise.