Discussing the Future of Intelligent and Autonomous Systems with Weilliptic

GM GM. Welcome to what's going to be another fantastic episode of DevNTell. The second one this week, in fact. If you didn't know, DevNTell is a 30-minute podcast held every week allowing founders, hackers, and anyone in between the opportunity to come on the show and showcase what they've built. Today, I'm really ecstatic to welcome Avinash Lakshman, who is the founder and CEO of Weilliptic. If you didn't know, Weilliptic is looking to create the most advanced decentralized infrastructure network ever built, providing unparalleled scalability without sacrificing security and throughput. If you stick around for today's episode, you'll get to meet Avinash, learn about his experience building some of the most heavily used DB tech in the industry, and learn all about his newest venture, Weilliptic. All right, let's get into it.

GM GM. Welcome to the show, Avinash. I'm excited to have you on today, man. You have quite a bit of accolades attached to your name in terms of some really fantastic tech widely used in the industry. It's really awesome to see you diving into the world of AI and blockchain. I'm really, really looking forward to today's interview. If folks might not be familiar with yourself, did you just want to give a brief introduction?

No, absolutely. I've been around the block for almost three decades, so I'm going to spare you three decades of history. I've been a systems guy for the larger part of that period of time. I was one of the three co-inventors of Amazon Dynamo, which became the genesis for the entire NoSQL movement, and then joined Facebook early on. Apache Cassandra was my brainchild. After my gig at Facebook, I did my first startup, Hedvig, where we built a distributed storage system, which I believe was the first of its kind, truly multi-cloud and hybrid, invisible kind of infrastructure. I did that for about seven years before we got acquired by Commvault for 225 million, and then stayed with the company for about three years. I got to learn a lot. Had a fantastic experience, I must say, and then left Commvault around October, November 2022, and formally launched Weilliptic, summer of 2023. And here we are.

Indeed, here we are. And yeah, I guess before we kind of dive into some of those experiences you mentioned, how did you just generally get into tech? Like, what really interested you to join the tech industry?

Wow, you're taking me down memory lane. I might not even remember some of those things. You know, to be really honest, I was a mathematician by trade. I was in the PhD program for pure mathematics, but then got into tech. I used to be extremely nervous of programming and things like that when I first started out. I always tried to avoid it up to that stage, but somehow things took a turn for wherever it went to and got into programming. That's how the whole journey started. And I always felt that, you know, I was not good enough. So that kind of kept pushing me to go and do things that I landed up doing, and it was mostly done to kind of reassure myself that I could compete with the best. I don't know if I've proven that yet, but that's what keeps pushing me in that direction.

I mean, some of your inventions and co-inventions are some of the most heavily used technology in the space, so I would say you've done quite well.

I will say I'm a sum total of all the people that I've worked with. So that has really influenced my thinking and got me to where I am. Some of the experiences have been absolutely fantastic, where it was really collaborative where people, especially on the Dynamo system, worked and learned together. Those are the kind of experiences that have shaped whatever I am today.

100%. Whatever you do in life basically acts as a little brick in the big foundation that you kind of set up for the rest of your career and life going forward. So yeah, I 100% feel you on that. And I guess since we're on the topic, did you just want to dive in a little bit around your experiences building Amazon Dynamo and Apache Cassandra?

Absolutely. See, a lot of people might not know this and even I might have forgotten some of the details because of time, but the current incarnation of, or I would say, the incarnation of Dynamo that the world is familiar with, happened by accident. Actually, the initial design that we came up with was shot down because of one question somebody asked in like a presentation. And then we had to go back to the drawing board to come up with what the, you know, the incarnation that the world is familiar with. That's how it came to be. So you could say, you know, we didn't know what we were doing and we landed up there. And that's how Dynamo came to be. It was a very small team and that was the first kind of system, I would say new generation system that came out from Amazon at the time. It replaced Oracle for the entire shopping cart experience. And then I moved down to the Bay Area more for family reasons. My brother had, my nephew was born. I wanted to be close to the family. It was easy for me to do that because I was single at the time. And then I joined Facebook. And Apache Cassandra, actually, I started working on Cassandra even before I joined Facebook. I started doing that as a pet project on the side. Never really knew where it would go, but there were certain things, my learnings from Dynamo that I wanted to kind of do it in a different way or do it better. I'll give you a simple example. In Dynamo, if I remember right, at the time we used BerkeleyDB as the underlying persistence, but then I decided that I'm going to go and write my own persistence engine. And these were things that I did more for my own learnings, you know, to push the boundaries of what I know rather than anything else. And that's how it started, to be honest. And then I joined Facebook and we found a nice use case for it. I worked with someone, a close friend of mine, who found a nice use case for it, and that's how we started building it out. And the rest, as they say, is history. Although I will say that a lot of people have worked on Cassandra after I open-sourced it. I've been kind of disassociated from it for a while now and a lot of people have taken it to greater heights, to be honest.

Yeah, but I mean still quite the foundational piece. It seems like all those experiences and working with similar like-minded people kind of, like we mentioned, built that foundation for what you did next, right? So I'm just curious, like what drove you from going from big tech and founding your own company? What was the inspiration there?

So, see, I always try to solve problems. I like to take on bigger problems. You could say this is a blessing and a curse. The blessing being that I'm always driven by complicated problems. If you give me easy problems, I'll probably suck at it. That's probably why I wasn't very good as an academic. But when I looked at the space and when I looked at the commoditization of hardware, etc., etc., making its way into the enterprise, I felt that something significant could be done and that's why I went on the Hedvig journey. And every system I've built, and I take great pride in this, I never solve the same problem twice. I like to use the learnings from the problems I've solved before and apply them, of course, but I want the problem that I'm going after to be a new problem and a bigger problem if I had the opportunity to do so. And with Hedvig, the system that came out, to give you a sense of the complexity, you could say that Cassandra was one of the modules that went into the larger system that we put together. So you can imagine how big that could have been, right? And but that was the reason why I founded the company, because we wanted to bring commoditized storage to the enterprise, where things like being location transparent, being able to address hybrid cloud, multi-cloud seamlessly without the user being aware of what is going on under the scenes and providing that seamless experience, I felt was a problem worth solving. And that's how the whole thing came to be. And most fundamental innovations I believe, for the most part, happen outside of big tech. Yes, they do happen inside big tech, but they're typically slower. There's a lot of, quote unquote, bureaucracy that sometimes you might have to overcome. And doing it this way, at least from solving the technological problem is easy, but there's always the business problem that you need to solve, because at the end of the day, to the investors, etc., etc., they don't care about the tech. You're there to build a business, not to build a great product, right?

Unfortunately, yes. Yeah, that's how it works out a lot of the time. And yeah, sometimes you just need that freedom without the politics, so to speak, to be able to do your thing. So I guess this is a good segue point to get into Weilliptic itself. So what is Weilliptic? What would you say is your elevator pitch of it?

Oh, absolutely. The way I would present Weilliptic, the elevator pitch is kind of complicated. And I'll tell you why. Let me ramble on a little bit and then you will get the, well, if someone's going to tell me that you need to come up with an elevator pitch, so be it, but it's a bit more complicated than just condensing it into two sentences. See, when I left Commvault, I took a look at, you know, I wanted to see what was out there for the next problem that I could potentially solve. I took a hard look at what was happening in the space of blockchain. And I felt that all the blockchains that were out there were monolithic in nature, meaning they were just one large cluster that could span and reside all over the internet. And that is a fundamental impediment to scale and throughput. So the first observation I had was something significant could be done there. Second, another problem that I saw is data sovereignty was never being addressed by anybody. What I mean by data sovereignty is people should be able to kind of define jurisdictional boundaries for where your data ought to reside and the system should be able to intrinsically kind of make that happen. Third, if you look at cyber security, it was taking over the enterprises big time and for whatever reason blockchain and crypto became synonymous, doing a lot of disservice to blockchain as a technology itself. The realization that I had was perhaps a blockchain could be a PaaS platform on which applications could be deployed. These applications typically are known as smart contracts as you know, but you could push the boundaries of that even further and you could deploy entire applications on an infrastructure of this kind. What do you get when you see that? You get you're immune to cyber security attacks, you get data persistence for free, and you can get people to forget about infrastructure complexity and these are properties that are intrinsically given you by the platform. Auditability, data sovereignty could be solved if you architect it correctly. So these are some of the problems that I wanted to address. And I'll get to the elevator pitch in a second. But these are some of the problems that I wanted to address. Along the way, I also started looking at all the buzz that was being generated by AI, right? If you go even today to Google and if you Google blockchain and AI, you will see articles that go back to a decade, but no one's ever spoken about what's the right way to synergistically bring the two of them together. And I looked at this and I felt something significant could be done. And in a bit to showcase that capability, we built another offering called Icarus, which is a virtual AI chatbot, which runs on-chain entirely. Okay, and I'll show you all those things in a second. Now, what problem do all these capabilities address holistically? And this is where it boils down to the elevator pitch. If you look at an enterprise, you can categorize the enterprise into two groups, IT and Business. And there are inefficiencies across the board. Now let's look at what those inefficiencies are. If you look at IT, the job of IT is to provide technological solutions to business problems, but every time they tend to do so, they have to solve a whole slew of orthogonal problems which are dealing with cyber security issues, infrastructure complexity, data sovereignty, auditability, data persistence, and the fault tolerance of it, etc., etc. Let's flip over to the business side. Business would ideally like to self-serve access to data, data that is locked in systems like Snowflake, Salesforce, Databricks, what have you. But instead, the workflow that manifests itself right now: file a ticket, file it to IT or to your data analytics team, and they get back to you with something and then you go about stitching your workflows together. AI could be a magnificent fit over there and that's what we tried to bring in through the Icarus capability, which is: build a chatbot which is fully aggregated, which has an infrastructure on which it is running, you can leverage the same infrastructure to deploy your MCP servers, wire them all together, and provide natural language workflows to the business folks so that they can literally talk to their Snowflake or their Salesforce or their MongoDB or their, you know, Databricks systems and tie multiple workflows together to provide that get rid of those inefficiencies that exist. So, in a nutshell, that's the large problem that we were wanting to address and I believe we have brought that out pretty cohesively. I'm going to pause. I know I said a lot, but...

No, no. I'm glad you said a lot, yeah. That sounds like quite the undertaking and certainly a different intersection of blockchain and AI. Typically, a lot of the things that I've seen played with really have to do with either providing some infrastructure to run some jobs, infra hubs, or some crypto trading bot, so to speak. So it's really interesting to hear this alternative perspective and a new use case here.

Exactly. I mean, if you go and, like I said, I think you hit the nail on the head when you go and Google for it, those are the only kind of things that people talk about. I don't mean to be dismissive, but that's kind of kindergarten stuff, right? I mean, once you do that, why, right? But you want to think of the problems that exist in a data center holistically: cyber security, data sovereignty, compute sovereignty, current geopolitical environment, data sovereignty is going to become extremely critical. And those kind of attributes that you would typically ascribe to a system cannot be an afterthought. You got to kind of build it in foundationally, and I believe that's what we have actually done.

And yeah, I mean, I think it's a good segue. I know you had a little demo for us for a walkthrough. Did you want to go through that? Yeah, absolutely. Let's jump right in. And for those of you watching us live today, if you have any questions, please drop them in the chat and we will get those up for you.

I'm assuming you guys can see my screen. This is the dashboard of the WhaleChain platform. This is a globally distributed sharded blockchain. I'm going to just zon in here. You can see this is the global view of the actual blockchain. This is made up of different shards. We have one in the US East Coast, one in the US West Coast, and one in Asia Pacific. If you drill down into what is in the US West Coast, these are the nodes that make up that particular shard and this particular shard is its own self-contained blockchain.

Now, if you go in here and if you click on this applets tab, it basically tells you what are the different applications that have been deployed on this platform. Now there is we call them applets, you can think of a one-to-one relationship for example between an MCP server and an applet and you can see all the MCPs that have been deployed on this platform. Now, there's another thing that I wanted to highlight which I did not previously. If you look at the industry and if you look at the last 30 years, a lot of work has gone on into building driving efficiency into the data center. What do I mean when I say that? The number of cores in the server have increased significantly over the last 30 years and so people started thinking about how many VMs can I pack inside a physical server or how many containers can I pack inside a physical server? How can I use all the cores that are there in the most efficient manner possible so that I can reduce my power and cooling costs, my footprint of the data center, etc., etc. And I believe in the journey of AI, everyone started embracing Python, which I believe is a giant step backwards because Python has its place in the developer ecosystem, but definitely building systems is not one of them, because it's not even a JIT compiled language and it basically blows your efficiency argument out of the water. So, in a bit to kind of address that what we did, was we decided we are going to let people build quote-unquote smart contracts or applets in four different languages. So today we support Rust, GoLang, C++, and AssemblyScript. Basically, the only ask is when you build these they should be compilable down to a WebAssembly target. And all these languages have implicit support for it. So for example, if I were to open up one of these applets, you could look at how the interface for the applet looks like. And this is in a format that we call WIDL, Weilliptic Interface Definition Language, and I'm sure we will publish the link to the publicly available docs. And we wrote a WIDL compiler that can generate bindings for this WIDL in these four languages that I was talking about and all you need to do is to program the stubs that get generated as a result. So these are some of the applets that we have deployed on top of this platform. If you look at this, Icarus is one such applet and Icarus is nothing but our chatbot. So this entire chatbot that you see over here is deployed on the WhaleChain platform. When I say deployed, I mean the JavaScript, the HTML, the resources that make up different items inside the HTML, the web server, everything is deployed on-chain and served out of there. And I'll tell you what the benefits are of doing that in a second. But in essence, this chatbot you might feel like you know, what's different? Like what are the different things that we could do in order to make this better than what's available in the market today? So if you look at this chatbot, it can be operated in two modes: conversational and agentic. Conversational is you could ask any questions of it and the responses would come either from the data that was used to train the LLMs or it would even do a web search for you. So for instance, if I were to say what is the weather in San Francisco today? This will result in a web search and you will automatically get the results for yourselves. The other thing you could do in conversational mode is this is something that some of our customers use very effectively, especially in the insurance space where, you know, people have these claims document that get filed and people don't want to read these claims document in order to answer questions. One could easily go in and then upload these claims document like I'm just going to pick our technical whitepaper. Upload it and then you could ask questions off it because there is a vector database that runs on-chain and that can provide semantic search which we just leverage. So you could say, what is the role of the Senate in the WhaleChain platform? And it will basically semantic search happens and it provides you information about the role of the Senate inside the WhaleChain. So this is pretty standard stuff. I mean you would have experienced this with perhaps most of the chatbots that are out there. But where things start getting very interesting is when we start moving over to the agentic mode. Here one can come in here and kind of set up your chatbot to get wired up with different MCPs. And if you notice, our MCPs are these are some of the MCPs that we offer right out of the box. And we are talking about serious enterprise-grade systems, right? I mean I'm talking about Aurora which is a Postgres service that's available on AWS. We have one for S3, we have one for Snowflake. This is just a small snapshot into all the ones that we have. I'm just using these for the purpose of this demo. And I can even show you how to wire one of them up and how easy it is. So let's say I remove this Salesforce applet from here. All you need to do is to come over here, we already have this applet deployed on the platform. I will just copy the unique address associated with that, come over here, and wire it up as follows. And once you wire it up, you can ask natural language questions of Salesforce. Now let's walk through a few examples and I'll tell you some of the things that we learned along the way in terms of how one needs to kind of interface with the LLMs. Most of the stuff that we learned unfortunately wasn't available online either. So we had to learn through these through mistakes that we made and through observations. So let's start with a simple example. Let's say I want to query Snowflake. So here you come in and you pick a data source and I'll tell you when this becomes important. Here's a natural language query for Snowflake. This is the first flow, so you can think of this as a task and a task is made up of multiple flows. And I'll tell you why that becomes critical in a second. And then we can say the second flow, we're going to take this data and Confluence is a very important collaboration tool that is employed in the enterprise. So what we're going to do is query Snowflake, get the data, and then as a part of the second flow, take the data that you got from Snowflake, put it inside a page in Confluence, and I'm just going to let this run.

And when you do this, oops, that's not what I wanted to see. Let's see. I think maybe you forgot to add the data source in the second flow.

No, I don't need to do that. Maybe that was a bad idea. Let's look at something different. Don't worry guys, this is only a test cluster, so things will happen. This is very natural, the nature of live demos. Yes. So let's stick to Snowflake first and I'll show you what it does. If you do this, this should work. So you can see it makes... Oh, something's wrong. I think someone's screwing around with the cluster. I apologize for that. No, it's all good, all good. So let me see if I can make this work with something else. If not, I will have to apologize. Let's come over here, pick Aurora, see what happens here.

There you go. So here is a classic query that we ran where I just went to Aurora and I said: it's equipped with a database called Chinook and I just issued a natural language query saying: retrieve 10 album titles and export it to a particular file. Now that we know this works, let's work around it with a multi-flow thing. So let's say we do this: this goes to Aurora, the second flow is upload the results of this into S3, and then the third will be send me an email to the link where this has been uploaded. And let's see how this all works together. So you can see it's a multi-step flow for this task. The first is: query some information from my Postgres database, upload it into S3, and then send me an email to the link so that I can grab the uploaded result at a convenient time. So let's see if this works. So the first is a query to Aurora, it's gotten it, and then the second should be an upload into S3. That will go through, it gives you the link to the S3 URL, and then it should send you an email. So now let's go into my email because I sent it to myself. There's my email with a link to the document that can be downloaded from S3 and this should contain the results of your query. Okay, now that we know that works, let's flip over to this one tab that I did not talk about. This goes to audit logging. If you look into this, every action that you performed on the chatbot, by virtue of it being deployed on-chain, every action anyone performs is immutably logged. So you can get things like auditability very naturally built-in without having to think of it as an out-of-band service, which typically you would have to do if you use any of the other chatbots that are out there. So for example, let's click on this. This basically tells us the natural language that I executed, this was the SQL query that was generated for it, and this was the tool call that was made against my MCP server. Now it was a multi-flow task, so you can see even the next one which was an upload into this bucket in S3 and it can be tracked. This is the unique ID that is associated with the individual who logged into the Icarus chatbot so you can track that task. You'll know who did what when. One other thing that I want to highlight, if you look at this input data, so this is actually this data. Now, we have the ability to mask the data and not display it in the block explorer, so to speak, and we do it using cryptographic commitments. Let me give you a very good example of why this becomes very useful. Let's say you have Toyota, Ford Motor Company, and Michelin tires working off of this chain. You do not want Ford Motor Company to see what kind of deals Toyota might have signed up with Michelin. If you go to any blockchain on the planet, all this information is publicly available. And since it's large enterprises, everyone will know what their ID is. So they will be able to somehow reverse engineer by looking at the data that gets displayed what kind of deals that they might have put together. A very good way to mask this is to lean on cryptography and using commitments to kind of mask that information. So no one will be able to see what this is, but you could actually use this to immutably verify this information that is on-chain and know that this indeed does correspond to the data that has been logged on-chain. So these are some of the advanced techniques that we have employed for auditability that you're able to kind of see over here, and auditability doesn't become an afterthought. It's kind of wired in into the system, if that makes sense.

Interesting. So it's kind of similar to, for example, when you're making those cloud.md files for Anthropic's model and having that kind of separation of like, oh I want this particular module to be built this way. So you might split out your cloud.mds across different modules to basically not context overload the model.

That's a very good point you bring up and a good segue to what I would like to talk about next. See, when you brought up Cloud, most chatbots out there today have an ability for you to wire up MCPs, right? But the onus of developing the MCPs and above all, deploying them, meaning the infrastructure for it, the auto-scaling of it, the onus of that is on the consumer. That's one. In our system, if you use our chatbot, you just focus on implementing the MCP. You don't have to worry about the infra. You can just deploy it on the WhaleChain platform in a serverless manner. So you don't have to worry about infrastructure or the auto-scaling of it. That's one. Second, a lot of people underestimate the size of the context windows. The context windows are pretty small. This goes back to it hearkens back to the time where in the 90s you had computers with 64 meg RAM or 128 meg RAM. So you got to use them very, very efficiently. So some of the techniques that we built in order to take advantage of that is you don't want to send the schema of the database every time to the LLM because sometimes the schema can get so large that it can blow past your context windows. So what we do is while deploying the MCPs, you deploy the schema and we then employ semantic search on the schema based on the natural language query that you would have entered. That way you only send what you need to the actual LLM. And second, if the results of a query are really large, the chain infrastructure has an inbuilt in-memory file system where you could put that state in there and just pass pointers to that data to the LLM in order to do the upload what have you. So again, very, very efficient usage of the context windows. So these are some of the things that we have really thought through from an engineering perspective in order to put this together. And the way we have put it together, the consumer need not worry about all this. You use these tricks and you don't have to worry about context window sizes, don't want to worry about infrastructure, and since I am in the West Coast, all my data I know will be directed to the US West Coast pod, so data sovereignty is implicitly solved and that's achieved through a magic of DNS and anycast, and I also showed you how audit logging is implicitly delivered for you, which is not with any other chatbot. You'll have to worry about how to wire all that in. In a sense, what we're doing is the mantra has always been: just focus on what one needs to do and don't focus on how you need to put it together. Makes sense?

Amazing. Yes, indeed. And while we are still on this topic, Winnie actually asks a question here. How does Weilliptic ensure that AI agents won't deviate or act maliciously?

That's a very, very powerful and interesting question. Now, here's one of the ways in which, see, I do believe, let me digress a little and then I'll come back. I'll try to best answer that question, it might not be the perfect answer, but one of the reasons deploying on-chain becomes really, really valuable is you can think of these MCPs as smart contracts. So once deployed, you cannot manipulate them. Tomorrow, if we want to deliver this as an agentic library, once you deploy an agent on-chain, no one can maliciously change them because if you change them, the contract address changes and you will have to wire it up with a different one. It's immutable. So those kind of properties can really go a long way in kind of addressing problems of this nature. That's the way I believe I would think of it. I do believe that in the future, AI and the provenance of it, knowing which data it was trained on, how do you eliminate bias, all this I do believe in the future can be accomplished if you can actually drive LLMs also from chain. And you don't have to put the data on-chain, but you can put signatures for the data on-chain so that it can be verified. If someone says, this is the training set we used, you can actually go verify that that was the training set that you used. So a lot of such things could be built and this becomes a fundamental primitive or building block to enable that going forward.

Great answer, great question. Thank you, Winnie. Appreciate that. And seeing as we're kind of closing in on time, and for our developer audience, how can people get started with Weilliptic today? Is there like a getting started tutorial or something you'd recommend for them?

Oh, absolutely. There are two ways you can do it. One, the documentation and the tutorials for how to build MCPs on our platform, how to actually even build an application and deploy it on our platform, is all available in our publicly available documentation: docs.weilliptic.ai. Very easy to remember. You can find all that information there. If you go to our website, weilliptic.ai, there's a link on top that says Icarus. If you click on that, there's a bunch of videos done by some of our engineers that showcase how the different MCPs were put together and they'll walk you through the individual use cases really well. Short five-minute videos, very intuitive. It's actually very, very simple to put an MCP together because what we have actually enabled for one is focus on the task at hand, right? You implement it, deploy it, and you're up and running. Don't worry about infrastructure, don't worry about auditability, don't worry about data sovereignty. It's all delivered intrinsically through the platform. And of course, please reach out to me, DM me on LinkedIn. I'm very responsive. In fact, a lot of people always accuse me of saying that I accept any invitations that I get. I don't really care. I am a man who is I still consider myself like a clown engineer, so I will respond to whoever I can as quickly as I possibly can.

There you have it, folks. And Avinash's contact info is in the description of the podcast below and we will also have the docs there for you as well for you to play around. And I guess to close things off, do you have any words of wisdom or any type of advice for some of the developers who might just be starting their careers or maybe starting their own companies? Yeah.

You know, when you say words of wisdom, you make me very nervous. But what I will say something that has worked for me, again, see anytime anyone gives an advice, they forget it really depends on the personality of the individual, how you're wired up, right? What works for me may not work for someone else, but I'll tell you what works for me. In most of these kind of systems, the only way to figure things out is by doing it. Because you will get two, three different multiple perspectives, but you have to do it and you will have to figure out what works, what doesn't. And that's the best way to learn. And that is the one thing that I will absolutely kind of emphasize. And before I forget, I apologize for the glitch that you guys saw, but that happens I guess, that's my excuse. But those are the kind of things that you will have to deal with in the real world in any case, right? Anyway, I just wanted to say that was not something that I liked to see, but I apologize for that.

You'd be surprised how many people have gone through the same thing over all the episodes that I've hosted here. So this is a very normal thing and it worked out in the end, so it's all good. Yes, for sure. And with that, Avinash, thank you so much for taking the time out of your day to come chat with us. Weilliptic looks really fascinating and yeah, I'm looking forward to seeing its evolution over time, as I think it's a very promising project and really drives what's actually possible beyond just the native crypto trading bots and whatnot in this intersection of AI and blockchain. Thank you for having me. My pleasure. And with that, just want to wish everybody a very happy Friday, happy weekend wherever you may be and we will catch you back here on another episode of DevNTell next week. Till then, have a good one, folks. See ya. Thank you.