Cloudflare TV

Dial Up Motive

Presented by Gabby Shires, Dhravya Shah
Originally aired on 

Human-interest segment asking Cloudflare employees what their first Internet experience was and how it informed them joining Cloudflare. Dial-up modems, bulletin boards, punch-cards, Twitch, Twitter and more.

English
AI
Cloudflare Interns

Transcript (Beta)

Hello there. My name is Gabby and I'm the host of Dial Up Motive. And Dial Up Motive is a show where we talk about the early Internet and how the Internet shaped people's lives and careers.

But today we're doing something a little bit different.

If you are a longtime fan of the show, you know that we typically interview Cloudflare employees.

But today I wanted to highlight the uber cool, uber talented interns that we have at Cloudflare during the summer 2024 session.

And I think it's important because when you're talking to people who were born in the early 2000s, we're Internet natives.

So we have a lot different experience than someone who might have been born in the mid or late 1900s and discovered the Internet later on in life.

We were born with it, grew up on it, and used it to build our entire lives and our entire career.

So without further ado, I want to introduce you all to Dhravya.

And Dhravya, tell us a little bit about yourself and what brought you to Cloudflare as an intern.

Yep. Hi, guys. I'm Dhravya. I'm one of the interns here at the Austin office.

And I work on the AI products and tools that we have, the workers AI and function calling stuff.

And what brought me to Cloudflare was very interesting and like first how I came to US.

So I was something called like an iPad kid.

I had access to technology from a very, very young age.

And that led me to play a lot of games and watch a lot of YouTube videos.

And then I happened to stumble across programming. I loved it. Then I made a product on top of the Cloudflare services.

And that blew up. And then I came to the US.

Yeah. Oh, my gosh. So that's amazing. So Cloudflare brought you stateside.

And I hope that you're having a great time here in the States and that you've loved your experience so far.

And you are originally from Mumbai, correct? Yep, that's right.

Awesome. So and you're also currently studying what at Arizona State University?

I am studying computer science. I just completed my freshman year.

And like I'll just start my sophomore year in a couple of weeks now, which is exciting.

So yeah, it's amazing. CS degree is very fun. Yeah. And you mentioned that you were double majoring in something.

So can you tell us what that is and why you chose that double major?

I am thinking about like I just applied for a double major in computer science and maths.

It'll be fun because I have like because of this internship, I have developed a really good, really big interest for math and probability and other like linear algebra and those concepts.

And I've started to dive in deep in them by myself. And I was like, OK, if I'm doing it myself anyways, maybe like I should just do a double major because it will not only help me in understanding the world better, become a better programmer, but it will like help me get more into like foundational AI stuff, which is what I want to do.

Yeah. So I love that. I love that you're passionate about artificial intelligence.

And you're even like modifying your degree program or adding something on so you can be even better at it.

And that's something that I noticed about you is that you're really passionate about the things that you do and you're constantly building, you're constantly coming up with really cool ideas.

And you're a little bit famous on Twitter.

So you actually didn't originally plan to be an intern this summer, right?

Can you tell us the story of what your original summer plans were and then how you ended up in Austin in our internship program?

Yep. So I have been building a lot of stuff like in my free time, even in college.

And whenever I build something, I make a video about it on Twitter and Instagram.

I post it.

And because I keep sharing my journey like that, I keep building stuff. A lot of people became really good friends of mine.

And then I happened to like get a lot of followers in the last couple of months.

Now I have around 20,000 on Instagram and 10,000 on Twitter, which is pretty cool.

And one of these days in December, I made a very small, simple thing.

It was called dump.place. And that product, so I made another tweet about the tech stack.

And then it got DDoSed. It was also an open source project.

It got DDoSed and I made a video about it getting DDoSed and how I protected it using Cloudflare.

And to my surprise, Matthew Prince actually retweeted that video.

Of mine. And at that time, I was just like really happy. Cloudflare sent me some merch, but that was kind of it.

I ended up like booking my flights back to India.

And I didn't even think of getting an internship or something.

And I was like completely ready to come back. And then out of nowhere, I saw a tweet by the senior vice president of Cloudflare here, Dane.

Dane had like, I'm hiring for people.

And I liked that tweet and we got in a conversation. And that conversation led to me talking to Walkin, who is the head of engineering.

And that led me to this internship.

Yeah. So that's amazing. So just to recap, like this is a fantastic story.

And I think it's very Cloudflare. So something about Cloudflare culture is everyone is chronically online.

So if you want to get in contact with anyone at Cloudflare, hop on Twitter or Discord or Reddit.

But you built something and it got DDoSed.

Cloudflare, which for those of you who are not familiar, Cloudflare has a really great offering of cybersecurity products.

You built a thing that got DDoSed, built a solution to the DDoS using Cloudflare products.

And then our CEO and co-founder Matthew Prince retweeted it. And then Dane, who is our, I think he's a senior vice president.

I might have that title wrong.

Oops. But he's over emerging technology and incubation. So the AI products and some of our newer flashy products.

He reached out to you on Twitter or X.

And then that's how you got looped into Cloudflare to join our internship program.

But you originally had no plan. You're originally going to go back to India for the summer.

So that's so cool. And I think that really speaks to the power of the Internet and community.

And how, again, chronically online Cloudflare folks are.

Yeah. So I love that. And we're super happy that you're here. So thank you for joining us.

And you mentioned this briefly, but can you talk about the teams that you're on and what products you've worked on here at Cloudflare during your internship?

So I'm in the workers AI team, but I've also been helping with some other small things, like in workers onboarding team and stuff like that.

That's not substantial enough.

And I've also been very involved in some of the Zero Trust efforts we have been doing that Dane himself has been trying to get done.

So at workers AI, we launched this thing called function calling.

That was my intern project.

And along with the function calling, we also launched a small SDK that any developer can use to make it very easy for new developers to come in and make AI agents or very complex AI workflows with less than 50 lines of code.

And it's actually an open source package as well, which is amazing.

You can check it out at Cloudflare slash AI utils package.

And then with Dane, I actually... So one random day, Dane messaged me, hey, you want to help me out with a script?

Then we met.

And so it was a huge problem. Like there were a lot of scam and phishing links that were piling up.

And we actually cleared out all of them pretty much slowly, but surely, which made the Internet a much, much safer place and a much better place to be in, which is very exciting that I was like a major contributor in doing something like that.

That's amazing. Yeah. And I think that, you know, as a digital native myself, right, making the Internet safe and getting rid of some of these like phishing and scam attacks, I think that it's almost like putting a lock on the front door of your house, right?

For me, at least, like when there are cybersecurity threats, I view the Internet as a third space almost.

So yeah, I want it to be safe and I want it to be protected.

And I almost take it like very personally when people threaten the safety of the Internet.

So I love that for you.

And I'm so happy that you got to contribute to it. And speaking of function calling, Dravya, there's something that you shipped four weeks in.

Would you like to talk about that?

Yes. So function calling was one of the projects that I shipped really early in my internship because of the combined efforts of the entire AI team.

But when I first came into Cloudflare, I had no idea what I'd be doing, but I knew that I want to be on like some flashy, cool new product and I want to make something bigger.

So Dane put me on this function calling team. And by the end of first week, I knew exactly what to do.

By the end of second week, I had a prototype ready.

By the end of third week, we like got it done. And then I was just waiting for a back-end change to be done.

And like fourth and fifth week, we just released it and it got pretty good traction.

And we wrote a blog. And yeah, after that, it's been pretty cool.

Yeah. Yeah. And so you released embedded function calling.

So can you talk a little bit about what function calling itself is, which is like already very amazing and cool and futuristic versus embedded function calling, which is what you shipped about four weeks into your internship, which again, mind boggling.

Yep. So function calling is basically a way to make AI models do tasks or fetch live data based on commands that the user offers.

So for example, you know how Siri works. You ask Siri to do something and then it can get the weather or it can change the lights, stuff like that.

But that's like pre-programmed stuff for Siri.

Like there's years and years of effort put into making that work.

But function calling is a way to make AI models do tasks by forcing them to give output in a very particular format.

That output is then used to run some function.

And then that function result is given back to the AI to give a natural language response back to the user.

So this entire flow of first me saying something to the AI model, AI model generating arguments, function run, and the function probably is accessing a database somewhere.

And this entire flow might take a lot of time to run.

And that's why we don't see AI models and AI agents in our everyday life here because they have a huge latency built into them.

Embedded function calling kind of helps decrease our latency by co -locating the CPU, that is the function run itself, the compute and a GPU, which is the AI model and the same place so that they communicate really well.

Not only that, but embedded means that the developer experience is much better because you don't have to worry about making API requests or anything like that.

All you need to do is literally run the one function that we have provided.

It has all arguments.

It's completely composable. And the return type of that is exactly what you give to the user.

And it just works. So that's two big advantages, latency and the developer experience that is much better with the function calling stuff that we released.

So, and I just want to double click on that because I don't know if everyone will fully comprehend how amazing this is.

So there is function calling, which, like you said, it's allowing a large language model to get data from an external source and build autonomous agents, which is already just like, that's amazing.

That's an amazing feat of the 21st century and very interesting in the AI space.

But it's a little bit of a slow process, right? So latency means that something is slow because it has to go to all of these different components.

You said function calling is not good enough. I want to make it faster and I want it to be better for our end users, which are developers.

And so you thought of a way to use Cloudflare technology and infrastructure to co-locate some of the things that cause the latency that make it slow.

So you made it faster, which makes it more accessible to people so it can be a part of my everyday life and I can integrate it into my home automation system.

And you made it easier for the developers who might build some of these autonomous agents.

So, yes, amazing. And not only did you do that, you did it four weeks into your internship, which is, like, I don't even have words.

That is so amazing. So, yeah, thank you so much for contributing to workers AI with embedded function calling.

And thank you for, again, making the Internet a better place.

Like, that is a very tangible thing that you've done to make the Internet better.

So I love that. Thank you. Yeah, of course.

But Dravya, you are a builder at heart. I think on your website, the first thing it says is I build products that people love to use or something to that effect.

You have built something that I feel like is probably the most Internet native, make the Internet a better place thing that I've ever seen.

And it's called SuperMemory.

So can you tell us a little bit about what SuperMemory is and maybe give us a quick demo and go over the architecture for it?

Yep, 100%. So SuperMemory is an app that basically allows you to build your own second brain by allowing the user to add content to the app.

And then you can query it whenever you want.

So I'll just show a quick demo and not keep speaking. So this is my website.

And let's say I'm just surfing my website as always. And let's say I'm not Dravya.

I'm someone else. And I just come across this. And I want to remember this person because I might want to reach out to him when I want some content creator or someone who has made a Redis alternative or something like that.

So by surfing this website, I can literally just click this one button and I can save it to any space.

Let's say Dravya space.

And then I can add it to SuperMemory. And in very quick, in just a second, it'll be done.

So this website is already added. And then on the app itself, you can go and you can actually ask about my profile.

So who was that one person who made his own Redis implementation in Go or any question like that.

And it'll actually find the information from my website. And there you go.

So you can see that's useful for an everyday person's perspective. And all the memories that are saved actually build up over here.

And this one single feed from all your platforms.

So let's talk about three big use cases. First is everyday people.

Let's say there's a teacher and the teacher wants to show the students. They want to make a collection of learning resources for the students.

They can do that.

They can put all the learning resources in one single feed. And let's say AI learning resources.

I've added some workers. And then they can just share this entire space to their students.

And then they can query this to learn about these websites.

They can talk to these pages. Other than that, this is also helpful for developers.

So let's say a developer looks through a lot of documentation every day in their daily lives.

And they might just want to go to a page like embedded function calling, let's say.

And they might just want to save this and then query it someday later.

That's also completely possible. So this is done. And then you can ask super memory like what is embedded function calling?

And then it'll find the page itself.

It'll also give me everything that I need to know and also the link to it.

One more thing that's my favorite feature is finding content that you've already been collecting since years and years being on the Internet.

So we made a Twitter import tool.

So you can just go to your Twitter account and you can click on this button in the bottom right of the screen.

And then you can click this button to import all your bookmarks.

Then you can talk to your Twitter bookmarks.

Like the other day, I remember saving a lot of full CSS tricks and tips.

But I don't remember. I want to find that person who did that, but I don't remember who it was.

So I just do, can you give me some CSS tips? And it'll find me the tweets.

And there you go. How cool is that? It gave me the exact trick itself. It gave me the Twitter handle of that person.

And now I can maybe reach out to them.

So yeah, that is super memory. Yeah. Yeah. So that is amazing. So I love that you went through the different use cases, because I think that when we have tools like this, it kind of gets like boxed into, oh, this is something for developers or technical people.

But like you said, if you're a teacher and you're like, yeah, I'm going to save a bunch of lesson plans about sight words.

And you want to look up like, hey, give me a fall theme of sight word activities for my first grader.

That's so powerful, right?

Because teachers, I mean, that's something a teacher can spend hours doing.

And my mom is actually a first grade teacher. So that definitely hits home with me as a use case.

For people who maybe they're someone who it has like a very specific eating plan, right?

And they need to eat like specific recipes.

And they've saved recipes from across the Internet. They can use super memory to look into them.

So yeah, it's giving Pinterest. It's giving chat GPT.

It is amazing that you built this. I absolutely love it. Can I ask what language model are you using for super memory?

I'm actually using a bunch of different language models.

So we are using vectorized to actually store the information inside super memory.

And then we are using GPT 4.0 along with workers AI models.

It switches based on the context to actually give the response to the user. We use Lama3 and workers AI.

It's super cheap. When we use Lama3, it's like 90 times less cost than if it goes to open AI, which is pretty cool.

So yeah, the entire tech stack is completely built on Cloudflare, even the database.

That is so cool.

And I believe that you have maybe an architecture visualization of how it's built.

Would that be something that you'd like to share? Yep. So one more thing I wanted to share is very important.

Super memory is actually completely open source. It's on super memory, AI slash super memory.

This is the domain. So it's a completely open source and it's free for you to use and see.

Anyways, so here's the architecture diagram.

And how I like to explain it is there's three big components that handle super memory.

There's the web app. There's a worker, like which is written in Hono.

And it's like an independent worker by itself. And then there's the extension.

The web app is a pages project and it's deployed on Cloudflare pages.

It uses D1 as a document store. The D1 is also really good because it's super duper fast and because it's SQLite.

So when the user comes to the super memory, that's what they see, the pages project.

When they add content, the add request goes to worker.

The worker has access to queues, KV, vectorize for caching.

It can get content from websites using browser rendering. It uses worker's AI to actually generate the responses and stuff like that.

The extension is just like the thing that you saw.

It's a small utility that kind of has an option to save batch tweets using queues or just save page directly to the D1 store.

That's amazing.

So completely built on Cloudflare. It's open source. This is just amazing.

The more I learn, the more I'm loving super memory. Can I ask you, with all of these different services and products that you're using from Cloudflare, with some other companies, like adding all this stuff in can get pretty expensive.

How much does this cost you per month for all of your users? Right now, I have 1000 active users.

And surprisingly, it's costing me, infrastructure is costing me about $5 a month to host because I'm just on the worker's paid plan.

And that is more than enough for me to run an app as big as this right now.

Wow. Yeah. So $5. So to put that into context for people maybe who aren't in the US, $5 is like, if I go to Starbucks and I get a coffee, it's about $6 or $7.

So for less than the cost of a coffee from Starbucks, you are running your app with 1000 users on Cloudflare.

Yep. I love that. That is so good. Yeah. No.

Wow. Super memory is so amazing. And I think that this is, when I envision AI and products like direct-to-consumer products that utilize AI, I think of something like super memory.

It feels like it has a great use case. It's easy to use and it's intuitive, right?

It's like I'm saving things then I'm going to query the things that I've saved.

And I truly believe that my mother, who is a 60-year-old school teacher in Ohio, would be able to sign up for this and use it.

So I love it. Yeah. Yeah.

In terms of getting started using Cloudflare, right? Because you, I mean, you're 18, you just finished your freshman year.

This is like your career is just leaps and bounds, you know, further ahead than mine was when I was 18.

But for someone who wants to get started using Cloudflare, where should they even begin or what products should they look into?

I think the best way to get started with Cloudflare is this Create Cloudflare CLS.

So just run npx create Cloudflare and this is the npm package.

This is literally the best thing because it'll ask you which framework you want to use, what you want to build.

And you can just use your favorite text style.

You can even write in Python if you want. Like even a programming language is not a concern.

You can write in Golang, Python, whatever you want.

And it's super, super duper cool because like from whatever background you come from, Create Cloudflare is a great way to start.

Oh my gosh. Yeah, I love this.

So this is amazing. And speaking of creating and artificial intelligence and Cloudflare, because Cloudflare, in case, you know, people weren't listening up until this point, has an amazing offering of AI products.

In terms of what you see artificial intelligence doing for society, what are some of the things that you're most excited about for AI to solve?

And you actually mentioned a project that you were a part of in Mumbai.

So would you like to talk about that and how AI might have helped you previously or might help you in the future with that project?

Right.

So when I was in Mumbai, I was actually a part of this charity. And we found a lot of underprivileged children who were affected by the COVID wave.

And they did not have a good education.

So they were set back by multiple years because of that happening.

And that was really bad for them, because when you devoid a child of education, that can lead to multiple generation suffering because of that one wave, even if it's a couple years of education.

An 8th generation child that I taught did not know simple multiplication and division, which is really sad to see.

And I think that is the one big thing that I think can be solved like education in general.

I was in actually a very good metropolitan city. And Mumbai is very, very lively.

And it has a lot of people. And I still saw this. So it makes me wonder how this problem, how bad this problem is throughout the world in multiple cities, big and small.

I think like just if everyone, one of these had some access to technology and a way for them to learn in their own language and their own style and their own way of teaching and culture, that would be like kind of the perfect AI tool, in my opinion.

Yeah, no, I love that. So thank you so much for the work that you did helping children get access to education.

And yeah, no, it's a huge problem.

So my mother is a school teacher in Ohio, which is in the States.

And there are a lot of students who have just skill deficiencies, especially with COVID and not having access to in-person classroom time.

There was a lot of skill deficiency that happened out of that.

Also, my video is frozen.

Let me try to unfreeze it. Sorry. But yeah, but that being said, AI, I think is a great tool because it allows you to make really individualized education plans, right?

In a way that we weren't able to before, right? If there's one teacher and there's 30 students, you're not making an individualized lesson plan for each one of them.

So again, education, I think is a really great use case for AI. And I'm happy that we have someone like you tackling it.

So beautiful. Dravio, what's next for you?

You're going back to ASU in the fall. You're going to complete your degree.

And then what do you want to do with your career? So at heart, I'm a product builder, right?

I want to make stuff that people love. So I'm not sure what I'll do after in my career, but whatever it is, whether it's me continuing at Cloudflare or in another company or starting my own company, it will be making products and it will be making stuff that a lot of people use in their daily life.

One of the biggest gratifications for me just while coming to office today was I spoke to my cab driver and he really loved that memory and he was so happy with that product that he signed up on the spot.

Stuff like that makes me just feel so happy that I want to keep doing it forever.

I love that. So I'll say this, Cloudflare would be ecstatic to have you, but no matter where you go or what you build, I know it's going to be absolutely fantastic because you're very talented, you're very passionate and I wish you the best, but I do hope that our paths cross again in the future.

Yeah, me too. I've been loving my time at Cloudflare so far.

It's my dream job and I'm like having one hell of an internship. Nice, I love that.

And I want people to follow you online. So this is your personal site.

It is dravya.dev. If people are listening at home, please pause the screen or go to dravya.dev.com.

It has all of his socials. I will tell you, I follow you, Dravya, on X slash Twitter.

You're a hoot. I'm constantly commenting on your posts, so you're probably sick of me.

But you are just an absolute pleasure. We're so happy to have you here and thank you for being on the show today.

Thank you so much for doing this.

I also had a great time talking to you. Of course. Thank you so much and I'll talk to you all later.

Thumbnail image for video "Dial Up Motive"

Dial Up Motive
Human-interest segment asking Cloudflare employees what their first Internet experience was and how it informed them joining Cloudflare. Dial-up modems, bulletin boards, punch-cards, Twitch, Twitter and more.
Watch more episodes