Ayush Ranjan I Building DePINs with Less "P" I Huddle01 I Pragma Singapore 2024

[Music] for our last Talk of the day I want to bring on our final speaker aayush to talk about what's happening in the world of deepin and especially kind of how do we go design more deepin protocols without the physical part in there so please give aush a big round of applause from huddle [Applause] hey everyone uh great to meet you uh I'll just check the slides okay great awesome I'm aush I'm one of the co-founder and CEO of huddle 01 hudle is building the world's first uh decentralized physical infrastructure Network for realtime communication now what's realtime communication RTC is anything what we do right now any kind of Google meet calls which you do any kind of Zoom calls which you do Microsoft team called Chu when you chat with a GPT uh or with any any llm via audio or video via voice or why video commands that's where you use RTC and that's what we have been building for last four years now uh right now I just basically want to talk about deepen in general and how in deepen uh realtime communication plays a very very crucial role and that's where I just basically want to categorize on where in deep in we are right now and basically you you'll be seeing a lot of essentially different kind of blockchains there different kind of compute providers there CDN storage and lot of other things huddle 01 comes under the category of RTC which stands for realtime communication uh this is something which we pioneered four years back and now fortunately after a couple of years we have been seeing uh that now in the bandwidth category RTC has its own subcategory uh which in itself is EV uh there are couple of other players in the space who have been doing some fantastic work including DOA Telecom and others which basically is trying to make sure that the way RTC has been built so far is built with a high operational cost right now if I just talk about the numbers we pay uh if you're using a zoom call you're paying essentially $20 per month per license and the reason why you're paying such a huge cost is because AWS has uh because Zoom has to pay a huge amount of money uh to AWS in terms of agress cost agress is nothing but the amount of data which you put onto Zoom or Microsoft team calls while doing a call just give an example if we all are doing a call we are a set of 50 people here and if we are doing a call here right now for just 1 hour we'll be uploading as much as 5gb of data onto AWS and this is just one room imagine the number of rooms which exist on a daily basis and imagine the amount of essentially uh money which Zoom is paying to um AWS in general now just giving you in terms of unit economics AWS pays to an internet service provider $1 but it charges Zoom $80 so it's an ATX multiplier just on the part of doing a routing the audio and video calls and that's where uh the the biggest barrier of cost comes into picture second is the geographical latencies the latency right now if you see if you're zoing that's the reason why we started H 01 back in the day where what we saw was that in the time of covid when anyone was basically doing a call an audio and video call and they were in a part of the country where there was no data center they had a huge kind of latency issues because all their audio and video calls if you're doing a call in Singapore was routing through a central server which is based in North Virginia imagine all your audio video packets routing from all the way from Singapore onto essentially uh a place which is situated back in North Virginia the amount of distance it travels the more distance it travels the more there is Jetter the more there is Jetter the more there is buffer and that leads to a kind of bad kind of quality of calls and that's the reason why if you see that we have not innovated a lot in terms of communication from the Skype days to the zoom days itself the scalability has increased but in terms of how communication could become much more immersive that hasn't happened yet the third is the Privacy aspect and the fourth is a value extraction aspect of things and RTC is a big Market I just wanted to basically touch upon this for a uh for a while because communication is everywhere we need communication all the things which we do right now but and these are the kind of category which is emerging and keep on basically going on it's something which is not very intuitive but if you see in anything we do uh communication plays such an important role here I just wanted to touch upon this right now but this is where I wanted to talk about my main topic of the day which is in deepen we talk about deep in is decentralized physical infrastructure Network which has a physical component mentioned into it now this slide which you are seeing here this has categorized essentially uh deepen into two phases one is a digit Resource Network where you just use a a digital Network and the second is a physical Network itself and if you see the categories in between them there are categories which include storage CDN compute which complet completely comes under the the part of digital part and then there is a physical Network which includes something like Bobcat minor of helium uh go.net Hive mapper and lot of others where there's a physical component uh into it and this has created a basically huge Fury right like completely a big kind of debate in the whole industry where it was started by uh one tweet which talked about that the rule of thumb is that deep in Project should have less P rather than having more p and that's where we realized and that's where there's a huge debate that hey you need to have more p uh more physical part rather than more digital part and the reason behind that a lot of these cases is that if you have more physical part it's more difficult to verify if there's more difficult to verify there's less scam which happens in the the network and you have a better kind of quality in general today I'm here to talk about that in building deepen it might sound counterintuitive but the Lesser P you have the better it is both in terms of scaling the network in terms of unit economics and in terms of the whole scalability in general uh and this is a basically a debate which has been prolonged for a while now and now it has been participated and been called as deepen as well as divin which is virtual infrastructure networks as well I want to talk about what we are doing at hle 01 to solve this we call this uh digital verification via qos uh which is quality of service if you see here how we have divided the way we are doing audio and video calls right now in itself or any kind of communication which happens we have divided this where we don't actively verify all these nodes basically right now how h01 solves this problem is that if V5 are doing a call and if you have a good internet connection you can become a node and you can power the call of all of us rather than relying on a data center which is very very far away from you now the question here comes that how do you verify such a exhaustive task and that's where what we do is that we say that we don't generally verify it in a very physical manner how we verify is that we do a rapid qos test all the time and that's where we have divided this subsection into multiple parts we call it registry we call it prover we call it Challenger where essentially the regist the prover and the and essentially which is the Challenger what they do is that these are the media nodes which essentially talk to each other and the the approver tries to prove that the task which has done is correct while the Challenger tries to essentially uh challenge uh the bit rates and all the other things so that's where the three kind of components which which comes here which is the bit rate that means the kind of uh how much audio video packets are you sending on on a per second basis that's first second is how much fraction loss how much of the seconds or how much of the packets have been lost while doing a call you would have noticed that when you're doing a call and if you're sitting in let's suppose um in any part of uh the country where the the internet not great you'll see that the quality of call degrades a lot and keeps degrading so it's it's basically a function of in the network who has the lowest kind of bandwidth and that leads to the whole call having not a great uh uh quality in itself so that's where fraction loss comes into picture and third is the latency latency means the round trip time how much time does it take for you for the audio back to deceive from my end to your end and this should be less than 100 milliseconds as this internal joke which we always have at h01 that uh uh that in blockchain sharding is one of the biggest problems uh but what we are doing by doing cascading that's even a bigger problem to solve which essentially means that making sure that you are having you're able to create this scalable infrastructure which has a lower latency than 100 milliseconds um and that to cross quently so that you can have a good quality call and that's what we have been trying to solve for 4 years now and fortunately we are here where we are now able to do the calls in less than 30 milliseconds where if it doesn't relying on a central server now you have these digital resource networks based on your own laptop uh I can talk about the specs moving forward but you just require a four core CPU a four core vcpu uh and you require just essentially 250 Mbps of internet connection and you're good to go I'll go more into detail of qos tests here there couple of things basically which we have uh the couple of simulations which we have run couple of essentially machination which we have run to make sure we have come to a essential score which talks about that how do you check that these qos tests which we do is not easily replicable that's the problem which every physical infrastructure network is trying to solve in a much more better way as compared to a digital Resource Network but that's where we say that digital Resource Network if the Qs test is done in a very methodical fashion includes Randomness as a function includes things like role rotation as a function includes all these things which make sure that there's no bias in the system you're able to basically provide the same and even more safer kind of verification uh out there which uh can make sure that there's no essentially game into your system in itself these are how we calculate the overall scores and these scores essentially lead us to if your score is more than 08 that means the whole network is performing great if your score is less than 6 then you basically 08 and 6 then it means that the net has some kind of game in the system by game what we mean is very simply doing an audio and video call and essentially the video is receiving you receiving the video late and there something which is happening with ch GPT as well right now CH GPT Bally partnered with a couple of players in the space and what they noticed was that when I am chatting with my llm and while chatting with llm the latency was super high like essentially uh it took me more than 5 Seconds to receive a reply of LM and that's not is the best way to communicate you require to have a super low latency so as we're moving away from the 2D screen environments of communication and moving to all of these new ways in which we're interacting with machines with drones with iots and with all these different kind of uh devices which means also with Healthcare and ettech where you require a HIPPA compliance so that there's a patient doctor confidentiality you require to have these kind of digital verification tests which is much more important and the best thing here is that this network can scale way more faster as compared to a physical Network because the logistics is a burden and uh you don't actually require that your unused compute your your unused storage your unused bandwidth we go after the bandwidth industry is good enough to solve the problem of communication in itself now I'll just talk about the couple of subsystems which we have built at hdle 01 these all subsystems are live for us deepen Network essentially means that you need to build the demand side first rather than building the supply side so right now if you want to use huddle 01 We have basically built the a kind of five step subsystems the first is is the client side the client side is our own app so you can use huddle 01 Meet app very similar to Google meet and zoom completely built using a mesh Network where essentially you can have a better kind of audio and video calls in general it's much more safer it's peer-to-peer in nature and you can store all these recordings over filecoin and ipfs as well making sure that your call Quality is good and also it's much more cheaper in terms of unit economics if you want to store huge amount of recordings over any of platforms and that's why we buildt the product layer first building this product layer made sure that we got social network effects we have now done more than 6 7 million minutes of meeting uh on the platform in itself and then we just compressed this basically use this whole network of ours or the app of ours and convert it into an SDK so anyone who requires Audio Video in their own apps in less than five lines of code and five minutes they can just integrate this app and they're good to go couple of app apps on lens protocol couple of apps on cyberconnect couple of apps on couple of games couple of apps on foraster ecosystem have started integrating hardle Z1 ASD into their own app because anyone who needs synchronous Network effects or synchronous communication needs to have audio and video and dealing with audio and video via just directly a bare web RTC is pretty pretty difficult in that sense now that's where our sdks comes into picture third part is our media notes which we all talked about that's the third subsystem here so the app side is the first the second the middleware is the sdks and the third is the media nodes which is essentially how we power this band within itself right now each one of us who are are here we essentially at our own homes have a 100 to 150 MVPs of internet connection which is good enough to power a normal call and that's what basically we wanted to show that you if you want to basically solve these hard problems we have and the way communication is improving in the sense of the internet speed both from starlink 5G moving to 6G you have enough Firepower at your own unused bandwidth which could be used to essentially make the communication much much more immersive much more much faster and having way better privacy as compared to any of the other uh uh centralized Solutions out there in the market um and that's where basically this whole subsystem works like you can right now out of the box use h0 1's app on a play store on an app store and can do a proper audio and video calls in itself talking about the network side of things which is our media notes it's pretty easy to set it up you can you just need to have these kind of bandwidth requirements as a minimum and then there's a bootstrapping reward to make sure that this is system skills completely in a good manner uh this is where the good part and that's where uh what we believe that every deepen should do uh is that build your own demand first that's one of the most important parts for building your own demand you need to have a good amount of product jobs and you need to have a good amount of app which you need to build yourself rather than relying on others to build the app on you because that requires a huge amount of escape velocity huge amount of product shops and huge amount of Investments to make sure somebody Builds on top of you the better thing to do is build the app yourself on your own network and make sure you're able to basically use that uh to drive Network effects to make others build on top of you and also build a much much more sustainable economy talking about here on our sdks since we already power a couple of apps now couple of games now couple of Health here now couple of ettech now and couple of others we are able to essentially uh uh do a total minutes of 1.2 million uh and also having a s good Revenue model here of $4 for every th000 minutes so anyone who needs to use our audio and video component just pay $4 and they're good enough to use uh r sdks on the client side we have our own app as I was mentioning done more than 4 million minutes of meeting uh more than 100,000 minutes of meeting have been done 100,000 meetings have been done it's pretty easy to spin it up you can just spin it up run it and use it on your Google calendar on your calendly on on cal.com very easy to use very smooth and you will feel the difference when doing the audio video calls you get the one80 quality of audio video and you'll uh you get a better codec and these things essentially becomes much more scaleout when you start basically doing communication away from the 2D screen to all through this more immersive kind of communication which includes multi-pc communication making sure that if you are doing a call in a VR and making sure you're playing a game with a you're playing a subway surer game with one of your friends now the more basically uh you're the more kind of uh uh movement you're doing the more there's multi haptic motions the more latency lesser latency you require so as we start moving from these communication Realms of 2D worlds to these essentially more immersive worlds we'll start seeing that the latency is such a big problem and how we require this people powered Network to solve this because otherwise the distance traveled would be so high so we need this essentially leads us to making sure that we are much more near to uh the brink of laws of physics so that the communication can increase better we have a better communication we have a better collaboration and we can have a better innovation in general uh now this is what hdle economy is users join the meeting uh apps is used they pay for RTC Services as they pay for RTC Services they pay that in the terms of uh in the terms of fees and then it's all being done by the media nodes where the media noes could be run by anyone having a decent amount of internet connection and a good amount of core CPUs now this is the ecosystem of hle 01 interestingly the apps here uh they all use us either on the sdks or on the platform side um so if you've seen the social F side these are the apps who already use h01 um uh on the from the SDK front the in terms of communities the one who who use our platform the reasons are myed the biggest reason for that is on the app side is because people are able to do video conferencing and live streaming at once this is not possible on any platform right now and then we are able to do that at a much cheaper cost uh think of it like Zoom plus streamyard Mor into one and you're able to do audio video call video conferencing and live streaming onto twitch onto X on onto quick and onto any kind of platforms as easy as possible um and this is why creators really like using this uh now this is essentially where we talk about what's next coming up so we have couple of things coming up in October we are essentially onboarding couple of our partners we are also doing our media notes which the white list of media notes is open that means anyone can now run these uh media notes themselves uh we're doing our white paper unhe we have been working on the white paper for a couple of um couple of years I'll say now and then our staking subsystems are coming up and then q1 we're doing our TG as well uh this is the kind of our our whole uh Journey so far so uh almost four years um and have infrastructure in more than 10 plus countries have done more than 7 million minutes of meeting um and now essentially growing fast uh this is our team our team is essentially of the 30 people based in us uh India uh Europe and couple of APAC regions uh we come from an RTC background we have done RTC for a long time now and being a huge proponent on why how the most important thing to solve in bandwidth is RTC and that's where if I just basically go back and talk about the the RTC component of it you'll notice that in this bandwidth Network map which you're seeing here there's VPN there's CDN there's AI data layer there's web proxy RTC and transcoding most interestingly if you're able to solve for RTC which means that able to have a way lower cross-continental 100 milliseconds of latency that's the hardest thing to solve in a bandwidth category and if you to able to solve that in a digitally proven way you're able to make sure that you can have a way you can solve any of the other things in the category so if you're able to solve RTC your network your network can solve for CDN a data layer and all the other things as well and that's where I'll just culminate my talk where any kind of deepen Network which is getting built to scale that Network out there are two things which is very important the first thing is that you need to build the demand side yourself uh to make sure that you're able to build a sustainable economy and second is that if you're able to digitally prove that uh and make sure that it's the Qs test is done in a very low latency manner you're able to have as hard to gain the system as hard it is to G the system in a systems like a physical device like hi mapper and a couple of other devices which is out there and yeah that's where I'll end my talk thank you so much and uh yeah that's about it [Applause] [Music]