Subscribe
Share
Share
Embed
Tractable is a podcast for engineering leaders to talk about the hardest technical problems their orgs are tackling — whether that's scaling products to deal with increased demand, racing towards releases, or pivoting the technical stack to better cater to a new landscape of challenges. Each tractable podcast is an in-depth exploration of how the core technology underlying the world's fastest growing companies is built and iterated on.
Tractable is hosted by Kshitij Grover, co-founder and CTO at Orb. Orb is the modern pricing platform which solves your billing needs, from seats to consumption and everything in between.
Kshitij Grover [00:00:04]:
Hello, and welcome to another episode of the Tractable podcast. I'm Kshitij, cofounder and CTO here at Orb. Today, I have with me Can. Can is the cofounder and CTO of Felt, which is a platform for making maps. Teams use Felt to make all sorts of maps, whether it's understanding weather patterns, graphing communities for environmental investigative work, or even in the software world for mapping out sales charts. So I certainly know I have a lot to learn about this world, so I'm very excited to dive in. Welcome to the show!
Can Duruk [00:00:34]:
Hello. Excited to be here.
Kshitij Grover [00:00:37]:
I definitely wanna talk a lot about Felt and the technical problems, but tell me a little bit about your background. How did you get into the maps world, and how does it relate to kind of the previous places you might have worked?
Can Duruk [00:00:50]:
Yeah. Sure, so my name is Can, spelled C-A-N. I'm Turkish originally. So, I came to the States almost 2 decades ago for college, ended up staying. And then since 2010, I've been in the Bay Area working in various different software companies. I worked in the past decade, almost 2 decades now, in a social media company called Digg, kind of a precursor of Reddit. I was there for a bit.
Can Duruk [00:01:14]:
I worked in a file systems kind of company, then I worked at Uber, for a few years then decided to take some time from the Bay Area, came back, and late 2020 started Felt with my co-founder and close friend Sam. How we got into mapping is two ways: Sam's previous company, Sam was the CEO or cofounder of a company called Remix, which you can think of as like Sim City but in real life. So he ran that company for 5 years and he got acquired. Sam was working on that space, transit planning software. if you've taken like public transit in most of the modern world, it's probably designed in Remix.
Can Duruk [00:01:52]:
We see people are using Remix's tools a lot to do stuff that is not transit planning. Or it's just you could once you make maps, it's like a computable, shareable, collaborative platform, turns out people wanna do a lot of stuff with them.
Can Duruk [00:02:08]:
But he is always thinking about that. I was at Uber for a little over like 2 years. So Uber, one of the obvious reasons for Uber is that Uber is not just automatically balancing equilibrium marketplace, there's a lot of internal tools and I would say a good chunk of those internal tools are like mapping tools. I was exposed to a lot of this “oh yeah, this kind of mapping, like model this city this way, model that city that way, you know this way”. So I was exposed to a lot of these mapping tools. So late 2020, both Sam and I were sort of like free agents. We were talking about some ideas we could work on together.
Can Duruk [00:02:48]:
And it was also almost like a distillation of like a couple ideas like we wanted to work in climate, you wanted to work in air quality, all this kind of stuff. It's like you just constantly look at maps. You might remember those like orange sky days, COVID. So constantly looking at maps, we're all okay, like remove everything from this. Like why are using maps so hard to make? So that was sort of like the idea. We're oh, yeah, we've done this before. I built mapping stuff.
Can Duruk [00:03:15]:
I was exposed to mapping stuff. It's not “that hard”. So we're okay, let's just build a mapping software because we know that everyone and their father and mother uses them, like energy, sales territory, newspapers. Anyone that you can think of is looking at a map at some point in their days. So we're “Oh, yeah. Let's just make mapping like the first class that is on the web”. So that's how the idea came around.
Kshitij Grover [00:03:41]:
I think that's interesting because I think a lot of people's experience with maps, sometimes it's in the context you're talking about, whether it's during COVID looking at all of these, these maps around outbreaks or even just air quality. But I think, primarily, people look at maps when they're doing kind of solo individual stuff.
Can Duruk [00:04:04]:
Right.
Kshitij Grover [00:04:05]:
So, I'm on Google Maps. Maybe I'm planning out my road trip. Right? So tell me a little bit about how the product works because I think you have this thesis that maps, of course, are much more extensive in terms of their use cases, but also they should be collaborative and they should be, in some cases, presentational. So tell me a little bit about Felt. Who's buying it? Who's using it? What are some of the use cases?
Can Duruk [00:04:29]:
Who's buying it? Who's using it? Everyone. I would like to send it to everyone. I would sort of zoom out, a little bit even further. So, imagine, one way to think about maps, and I think this is how a lot of people think about maps is like a routing tool. you put in I'm going from my home to school, from home to work, you put in destination A to B and give you a route. That's why use cases in maps? I think most people think about it, taking the use case of maps as POI, like Points of Interest.
Can Duruk [00:04:58]:
You type in, like Thai food near me or, you know at the restaurant, you look at reviews, you look at pictures. But if you think about it like zoom out even more, a lot of the data that you think of as data, think of like spreadsheets, think of like databases. There is a geographical component to it, right? Let's look at your company, like your org. We have customers. Your customers look over different states, different countries. I get into various database table somewhere. “Okay you have this many customers in this state, you have customers in this country, and just so like tax jurisdiction”. So there's a lot of giant sets of data sitting on everyone's database, Excel spreadsheet, wherever, that has geographic component into this. And a lot of the time you wanna see that data in its most native formats.
Can Duruk [00:05:46]:
You wanna operate in its most like native format. And it turns out like maps is that native format. You wanna be able to make your database, drag that file. So you'll see a lot of, let's say, like an energy company. They wanna understand, okay where are the power lines? Where are sort of combustible stuff around these like power lines? Okay. Like where are these power lines gonna go? Like where is the energy gonna come from? Okay. it's just somewhere. Like what's the, like the demographic, like what's that demographic look like? That's cool.
Can Duruk [00:06:15]:
If you go to a Whole Foods' website, there's like a place where it actually submits, “Oh I want a Whole Foods in my location, I want Whole Foods in my neighborhood”. And then you'll see what the end of the day requests from you. They'd be “Oh, like send us like data about like food traffic, send us data about sort of like population density, send us data about like demographics what is sort of like the average income level”. So you can see every industry, the restaurant chain does this all the time because they want to expand. It has more restaurants than anyone else in the US. So every single industry, there's like giant slots of data that are geographic components, and that there's sort of a way to immediately work with that kind of data today.
Can Duruk [00:06:56]:
I think that's where that's the gap I think Felt is trying to bridge.
Kshitij Grover [00:07:00]:
And it's interesting because I think the way that any given industry wants to express that data or display it is pretty use case dependent. So, one example that comes to mind related to what you were just saying is PG&E has a power outage. And then they send you this notification that says, “Oh, here's the highlighted areas where things are degraded or, know, there's maybe limited power, and and here's the regions it affects and that might be a very different sort of overlay or use case than someone trying to do journalism with apps”. So how do you think about addressing the broad set of use cases where some of them are very detail oriented and some of them are very kind of broad strokes presentational, you're sending it out to a thousand or maybe even millions of people.
Can Duruk [00:07:50]:
I mean that's definitely correct. This is sort of like one of the challenges of this industry, I think of it this way, this is not sort of like our company, necessarily per se, but I think of it as sort of like an Excel, right? Excel is used by everyone. And like you, Steve Ballmer uses Excel to manage his day, and I’m sure I bankers just Excel in a different way, and then you and I just excel in a different way. Right?
So I think you need to sort of build with Sam likes to call, you know a lot of good software has like depth. It looks like little simple stuff, but if you have the good primitives there, you can “hack your way to be used like other ways”. So one way we see Felt being used a lot is I don't even touch on this. It's like presentational. You know for example, people like working on sustainability, people working in these management consulting for like city governments or, specific consulting for like energy planning. They use Felt to essentially almost like a sales tool. They actually, let's say, they do a lot of quantitative analysis, but you need to like sell your stuff to customers. So like you have a consultant, you have a client, they use Felt for two ways. One is that they do internal collaboration because we have all the collaborative stuff with them. And Felt makes you look good. That's one of the key things that I'm sure, you've done enterprise sales. I'm sure at some point in your life, it's like making the person look good.
Can Duruk [00:09:25]:
The person using your tool looks good. So like a lot of people use that to look good. And I think that's where sort of like one of our key differentiators comes in. Sam is a designer. If you look at Remix, it has a sort of like love put into the product. Yeah. I don't want to say Felt is sort of consumer-y, but Felt sort of makes you look good because we actually pour hearts and sweat into making the UI snappy and like making you look good. Like a lot, like every single color tries to like work with every other single color, so you don't have to be a designer to like make yourself look good.
Can Duruk [00:10:03]:
Yeah. And once you do that, it makes it easy to use on the consumer side, but also if you use Felt as like a sales tool, okay I can put together some arrows there notes there and it's gonna make me look good too. So it's not easy, but I think we have some superpowers that help us be as horizontal as possible.
Kshitij Grover [00:10:23]:
Yeah. What that reminds me of is that the classic kind of sales thing in maps that I think of when you go there is the T-Mobile 5G network map. Right? It's like all these wireless service providers trying to convince you that they have the best coverage areas, and they cover every corner of the nation. So let's talk a little bit about those technical challenges because you mentioned a few things there. One thing is being able to really optimize for performance. But in general with maps, I imagine there's a lot of interesting technical problems. You have to think about how they work at different zoom levels and even things like adding annotations and arrows, notes, and uploading different data. I imagine there's a whole range of problems.
Kshitij Grover [00:11:05]:
So can you tell us a little bit about those? When you're building something like Felt, what is the starting point? What existing technology maybe can you borrow or at least start from? And then what have you had to, effectively build up from there?
Can Duruk [00:11:21]:
It's a good question. I mean, there's like two ways to answer this. One is patting myself on the back, one is sort of patting on the shoulders of giants. I would say when we started a company, it felt like there was nothing that we could do for certain things. Like we had to build a lot, especially drawing. So let's look at problems like Figma, FigJam. So they're like vector based drawing tools, but I know what they did is obviously, like it's incredible what they did, and you know we take it for granted now, but one of the couple things that you kind of like squints and I realize like there's certain things that make the problem slightly easier is the zoom level is limited. Like you can only zoom in so much Yeah. Figma or FigJam. But unlike a map, you need to be able to zoom in essentially, you're filming all the way to this, a continental level to like street level. So okay, what does that sort of like mean for like drawings? Do they disappear? Do they not disappear? I think that was sort of like how the drawing suite of tools. Like we had to, we used like a couple days that existed like none of those seem to work well. We had to build it ourselves. It's like a drawing challenge, like drawing tools. That's what we have to build. On the map renderer side, that's sort of like an OS level hard problem partly because there's just so much to do, partly because they just kinda run into browser issues. You kinda like to run into network issues.
Can Duruk [00:12:50]:
So on that one, we used to use, we switched. We've been using other renderers. Previously we used to use Proto Maps. Now we're using MapLibre GL JS. So now we stand on the shoulders of giants. When it comes to data processing, this is where we do both a lot of stuff ourselves and also stand on the shoulders of giants. I think the biggest workhorse is a tool called Tippecanoe, written by Erica Fisher. Tippecanoe, think of it as sort of like converting files into tiles.
Can Duruk [00:13:23]:
So tiling is like, and you can kinda like zoom, you can kinda like think about it as sort of like millions of heuristics like embedded. Let's say you upload something. It's like a famous data set. Like all squirrels in Central Park are all okay. Trees in Central Park. They're okay, what does that sort of look like at the zoom level? I like that zoom level? and it's a city if you're looking at the street level, you wanna see every single street. Or if you're zoom, you wanna see sort of like a blog. So there's a lot of heuristics, with you know There's actually several different things. So Tippecanoe is written by Erica Fisher, who works with us, for us. So now we maintain Tippecanoe. I've obviously doubled their other projects like GDAL, ogr2ogr, So those are open source projects that we use, but I would say a good chunk of Felt's stack is written by Felt. Partly because it didn't exist until we tried to do it. Partly because we need to control all parts of the rendering stack so we can kind of stick our hands in add hooks, so we can add more value to those products. So we own the entire stack. We own the stack that writes all the way that the base map, like the base map pixels are stored in our devices, and also like they're rendered by our renderer.
Can Duruk [00:14:52]:
So like anything in between, we had controls to simplify it, add value to it, remove value to it, add attributes, special case attributes, render attributes differently. So that was like the reason why we had to do a lot of stuff ourselves.
Kshitij Grover [00:15:07]:
And how much, if any of that technical decision making is informed by the broad applicability of the audience, that is, potentially even using these maps. Right? So, for example, unlike some other tech companies, you're not only selling enterprise sales, B2B to other tech companies. You presumably have utility workers using these maps maybe on their phone. Right? And so how does that inform maybe browser compatibility, slow network connections? What's the range of consideration there for the end user that might be looking at this map in an environment that's not just a Mac Pro sitting on their desks.
Can Duruk [00:15:48]:
Of course. Yeah. I mean, it's a huge consideration. You gotta meet people where they are. We're not selling this enterprise-only tool. I think people would like to send laptops to their end to our customers and be oh, your laptop is so old. So it's just like a thing that people had over the company that did this. Obviously, we can't do that. So we have to make sure that this stuff works everywhere, and it works really fast. So for it to work everywhere, we have this automated test that tests the product in like Chrome, Safari, and Firefox. And for it to be like working really fast, we also have automated testing both like on all PRs, some suites of tests that are run, and also like on daily. You have benchmarks, like FPS, various click around, like see how long it takes, and then you get alerted.
Can Duruk [00:16:36]:
So it is something that's top of mind for me. You have to do that because yeah, when we start a company, you gotta have you gotta have these talking points, right? One of the things I did was just I want Felt maps to load as fast as a screenshot, you know. That's what is like because like screenshots obviously there's cheating because they flatten PNG. But like I mean, it's not that hard. Like what we can do a lot to get it as close as a screenshot. So there's a lot of automated systems. There's a lot of processes, and there's a lot of culture to make sure that things work really fast.
Kshitij Grover [00:17:10]:
Yeah. And maybe you can talk us through that a little bit. So let's say I go and, maybe I go to a news article and within it embedded is a map built in Felt. What's actually happening there? Is there some smart loading of the map base layer and then the data embedded in it? How do you deal with interactivity in those contexts or maybe kind of been depending on how it's embedded? But, yeah, talk us a little bit through what actually is happening under the hood.
Can Duruk [00:17:39]:
Yeah. I mean, before we go into embeds are loose, embeds are largely iframes, right? So your embed is gonna be as fast as your map itself. So let's go into the map itself. There are a couple of things that we do. None of them are sort of rocket size, but you just have to do them, and sometimes doing them is not that easy. Right?
Kshitij Grover [00:17:58]:
Yeah.
Can Duruk [00:17:59]:
When you load a map, when you load a map, we stuff as much data into the first HTML response as possible to render the thing very fast. So obviously there's and I think that sort of like throws people off when we start working out loud because we're very interactive. Obviously, it's like 2D, like 3DS, like a drawing tool. They like to react to a split but like we actually stuff as much data into it as possible. So I don't like when you have that HTML response, you don't really need to do anything else other than of course, the base map tiles. So elements are into that. All the bounding boxes, all the center.
Can Duruk [00:18:39]:
So there's a lot of stuff that we stopped everything in and that I think catches people by surprise. The second thing is the base map tiles. When I say base map tiles, think of Google Maps also designed with the base map tiles. They call street tiles at times and then the satellite images. And the satellite images are just raster PNGs. Base map files are Vector. Vector, so they are super compressed and as well like a few months ago, they actually come from our servers. So we control the highlight pipeline where like the base map tiles come from CDNs that we own.
Can Duruk [00:19:11]:
So they're like a lot of stuff is processed beforehand. And the third thing is, and this is what makes mapping the cloud much much faster is, there's the class of data visualization that people do with maps, where most of the processing happens on the clients, which is definitely like a lot of benefits to it. You have got a lot of interactive free. But you really need to simplify data quite a bit for it to work. So we do the simplification on the server. So that's what Tippecanoe is, what I thought about trees. So we simplify data almost for every single Zoom level beforehand. So like every single Zoom level we're only getting from the browser, from our servers what you need to be getting and nothing else. So we're trying to limit as much as not send you everything but nothing more. And that simplification that happens on the server is writing one of our magic sauces that
Kshitij Grover [00:20:15]:
Yeah.
Can Duruk [00:20:16]:
you know you upload a 5GB file. People have this expectation that, like Excel, you wouldn't expect a 5GB Excel file to load instantly. Intuitively there's like millions of rows or millions of columns, you can have it oh yeah, like it just works. And it can work, but you need to sort of like MP3s as they remove all the stuff you cannot hear. Like on the map, like there's only so many pixels on a browser.
Can Duruk [00:20:45]:
Yeah. And then there's like at a zoom level, we know how many pixels there are. So we're okay, like at this pixel, at this density, we can only show this much data anyway. there's just like not anything else to stop it. So we do that simplification on a server. That makes it, I think, much faster. Obviously, on top of it all the interactivity we build ourselves, it has to be faster, that kind of stuff, but that's sort of table stakes.
Kshitij Grover [00:21:09]:
Yeah. And I imagine that one of the things you're dealing with is as an industry, the benchmark is a very heavily staffed team. Like people are used to this, like Google Maps experience where they've probably spent a long time trying to think about performance and the rendering characteristics of their map.
Can Duruk [00:21:28]:
So Google Maps, I think, is like a class of mapping products. Right? We talked about but you kinda look at how professional you use maps with tools, like maps based by like other companies like the industry behemoths and this, they're not good. Like I'm just saying this like finally. Because they become like you find these like maps, during the wildfires, right. You would see just like maps made by counties, like fire departments, like they would barely load. I mean you have that intuition, like you would go through the screen, it would just like, yeah.
Can Duruk [00:22:02]:
It would just refresh you. And like let's say you make this like a filter and you wanna share it with your friends, good luck. Because like all the filtering happens on the client is not saved anywhere. You send it to your friend and it doesn't load. It breaks the amount there. So that's, yeah, I think Google Maps, Apple Maps do an incredible job. But they are just loading the base map.
Can Duruk [00:22:24]:
They're not loading as much data. Right? Yeah. We work with giant sets of data. And on top of it, they allow me to do things like drawing and interactivity. And like that, the explanation people have is actually pretty low. And that's one of the things that we felt. People were oh, wait. We see a lot of people waiting, how does this work? And I mean, you just do the computer science thing.
Can Duruk [00:22:47]:
That works. I think we're used to seeing these aligned maps with giant sets of data being up to performance, and then we're trying to fix that.
Kshitij Grover [00:22:55]:
Yeah. And actually, let's talk about that kind of before and after a little bit because I think you're correctly pointing out that the sort of professional maps use cases aren't in these kinds of consumer maps applications. Talk a little bit about that technology. How did people upload datasets to maps before? What was the world like pre-Felt? And, and, again, even there, I'm sure there was a range of possibilities, but typically what did that look like?
Can Duruk [00:23:20]:
I would say, for the purpose of a podcast, I think most things like this happen on the desktop. Like that it just happened like a thick client. That's like most of the world we live in. So there are benefits to it, obviously clients have a lot of power. That's arguable, I think, in 2024, but compared to servers. But that's sort of like the world. The challenges are numerous. Obviously the first one is like a static program. If you need new versions, you have to download new versions.
Can Duruk [00:23:53]:
The biggest tool used in this industry, it doesn't have a Mac version. It's like Windows only, right? And, I mean, it doesn't even look modern like a Windows software, But it's very popular and like you can ascribe it's popularity to like whatever reason you wanna ascribe it to. But the second thing is that collaboration is like basically, stuck in back down worlds, you know. I doubt it is like immigration I deal with lawyers, it's like v1. Like it's been like 10 plus years that I've had to do this like ping pong or Excel or Word documents around, but like that industry still works with that. I think that's like moving slowly. There's like a lab version, but it's like you can kinda tell if there's still a share point level versus like, yeah. Google's next level.
Can Duruk [00:24:36]:
And the third problem is, I mean, I just think our software is better. Like this is, yeah, and I think the part of it is better that we can just keep training on this again and again based on user feedback. Right? Like we just like conflict. Yeah. So, the software gets there automatically. It's like a collaboration. And I would argue servers are much faster these days with thick clients.
Can Duruk [00:25:00]:
Like, having the word thick client obscures a lot of what can actually happen on the server if you align your data.
Kshitij Grover [00:25:07]:
And do you think it has opened up genuinely new use cases? Because imagine what you're saying where, you have to have this desktop application running on a specific sort of client feels like the setup for a specialized role at a company where someone is uploading a bunch of data, organizing the map, publishing the map, sending it to someone. Right? Whereas it really does feel like working on someone like Felt, working on something like Felt would open up the process of collaboration to a lot more people in an organization. Does that resonate? Is that a shift you've seen?
Can Duruk [00:25:42]:
Oh, yeah. Yeah. I mean, this almost feels like this is I like segue, but exactly the workflow that we see. What has happened is, I think for two reasons, like one you mentioned, a common sort of pattern that you see is there's like one person inside a company that knows how to use this GIS, Geographic Information System software. Right? That's sort of like the closest thing we're in. So that person knows how to handle that software and both because the software is hard to use then we clash on this. It's maybe, it's like it's hard to install and use. Not because of the UI.
Can Duruk [00:26:19]:
It is genuinely like thick software that needs to run on the power server. If you go to GIS Subreddits, there's like a monthly thread. People ask, "What's the best laptop to buy for my GIS tool?” Like monthly. Because it's like it's not an industry awash with money. People are really care about, like they really wanna buy thick clients, but the laptop they're gonna buy is underpowered. Like underpowered liquidity. So, and what happens a lot is because software is hard to use, you rely on one person, and that person becomes like a bottleneck.
Can Duruk [00:26:53]:
So if anyone has to oh yeah, like throw it over to Paul, and like I'm waiting on Jess. That's a workflow process that we wanna eliminate. And I think people get to add, marketing team. I can add an analytics team because it's online and the latest version is always there. There's undo, you can duplicate maps, so there's a little versioning.
Can Duruk [00:27:16]:
So it opens your company up to a different workflow and then as we add more and more features on the server, it's oh, yeah, like what was usable only by marketing now is also usable right now, like our the analyst because you can now like up on to like a few weeks ago on Felt, you can only see if you wanted to see raw data, you were kind of like limited, but we fixed that. You actually can see the entire raw data. You can actually operate on the entire like raw data, both on the client and the server. You like to process it. So it has become this much much thicker software that just runs on the server, but all you need is a browser.
Kshitij Grover: [00:27:51]:
And maybe it'll be interesting to dig into a little bit of the technical challenges that are ahead of you today and then maybe how they fit into the workflows. I think, in conversation so far, we've been talking about a few different aspects of Felt. Right? Here's the rendering of the map. There's the uploading of the data. There's the kind of collaboration and versioning, and maybe performance, we can say, is a bucket of its own. So, what does the team at Felt work on now? What are some of the technical challenges that you think are still ahead of you, and maybe how's that changed over time?
Can Duruk [00:28:26]:
Yeah. I mean, I don't wanna hold my roadmap cards close to my head. I think largely there's sort of the big thing I didn't like. There are two big things the team is working on. And you can almost file them into a bucket of work with bigger and more sets of data. It's both a qualitative and a quantitative difference. 1 is that the GIS world lives one of the things that we actually really are very broad. We also work with JSONs, but also work with Snowflake databases. So we're just constantly like adding more and more types of these databases that we wanna support, and sometimes the file type or database you wanna support is so different. You have to architect your system and fundamentally react with your system.
Can Duruk [00:29:16]:
So there's a couple of things really, architecture coming. One thing we're working on is sort of live data, which is right now. Felt essentially still requires you to upload data once and then, then why you can do this via your own API and we upload the data yourself. But ideally, you wanna be able to here's my database, Felt, handle it.
Can Duruk [00:29:37]:
And then, is that something like a technical problem? It's not like a computer science problem. If you squint your eyes, it's like an engineering problem. That's again one thing that we're working on. On top of it, on the more like the application level things, we found that like people like wanna work, people wanna embed the files, not in the Iframe sense, but like embed files in their work flows more and more. So the nice thing is that we have like customers and like they're hacking their way, like they embed files in their workflows. We're okay, like they'll do this.
Can Duruk [00:30:10]:
We'll, which is like a nice feeling where like we take like what they did to like hack together and I don't know if it's a word, but productize it. Okay. Like we actually made this like a real product. So you can see that it flows like an API. Okay. Like you did this weird little thing, we'll turn into the proper APIs so we can manage it, we can make it faster. And on top of it, there's a cornucopia of features, you know, people want more widgets on the map.
Can Duruk [00:30:35]:
Alright. Well, I try to figure out where they're going. People wanna have more localized ways to work with data. Oh, I want to localize this version. You can kind of think of also, obviously we're in the business of data. People are oh, where's my data going? Is it in the US? Is it in Europe? Is it in Asia? Can it run on my own servers? And now, people like answers to all of them. The question of sequencing okay, in which order do we attack these problems? But to answer your question, adding more data supports, both in terms of larger datasets and different data databases, and also allowing files to be embedded in your workflows by better API. I think those are the two big buckets of work.
Kshitij Grover [00:31:19]:
Yeah. And one through line, I think, throughout all of those things, but even what you already have in Felt, to me, seems to be there's this question of how configurable do you make things. Right? So visual density, even the kind of colors of the display, the thing you just described, people might have very different requirements for how often data refreshes if you're connecting to a live data source. What do you think about that problem? Because, again, with this idea of serving a broad range of use cases, I think there's definitely some value in being opinionated and kind of pushing the industry forward almost. There's also, I imagine, just some natural configurability you have to allow to serve the use cases. So what does that look like internally or even just in conversations with customers?
Can Duruk [00:32:03]:
Yeah. I mean, that's a good question. I think Sam, my CEO, is a designer who leads products. He calls it like software has depth, like that's what he means. It looks simple, but like there's actually more to it, you know. I grew up around cars, so I think you can open up the hood in Felt quite a bit. It's not like a Mercedes electric vehicle, but it's like a Toyota truck. Like there's a lot of things we can do.
Can Duruk [00:32:27]:
For example, we have a visualization language called FSL and then there's like an editor, you upload like a giant sense of data. You can kind of figure out how it's categorical visualizations. You can raster algebra if it's like a raster file. So there's a lot of stuff that you can do on the UI. And there's also, you can actually call it literally we have source and actually made them, made the button say we have source. You can click a button and like to see how that visualization is built. You can literally see some of our internal source code and change it yourself. So yeah.
Kshitij Grover [00:33:02]:
That's cool.
Can Duruk [00:33:03]:
It's really quite interesting because obviously, we started being the engineers we were and now we wanted to change the language ourselves quite a bit as we're working on. So we started with the language ourselves, our style language, with readability in mind quite a bit. Because we want it to be easily readable, all that kind of stuff. And then we wanted to rebuild a code editor inside, and we call it advanced editor. One of the things you immediately see is people like freak out the moment they see a curly bracket. They're oh, no this is like magic incantation. I will screw this up. So we're okay. We kind of slowly hit that, but it's not done.
Can Duruk [00:33:44]:
I think one of the things I've been learning is that more and more people want simpler things, but it's probably okay to start with the advanced stuff and simplify it as people want more. Like say I need I'm, I'll like to share some parts of the road map. We like API visualization. Like you can visualize the files map entirely through the API. It's there. It's possible. But turns on a lot of the time, even like developers who are close to themselves are enlightened beings that cannot make sense. A lot of time they want is can I send an API request that makes everything yellow? Yeah, write that JSON.
Can Duruk [00:34:27]:
Like that's not a good answer. The good answer is where's an API point that takes in like that kind of request, like we do the hard work. So I mean that's feeling like a journey. We build a lot of advanced stuff and then some people use it, but even the advanced people that's a spectrum. Right? Like someone's advances, like someone's simple. So we've been adding more and more simpler ways to do it while keeping advanced stuff around.
Kshitij Grover [00:34:51]:
Yeah. It sounds like what you're saying is you started with the primitives, and then I'm I'm sure you've iterated on them a little bit, but now you're, I think, sensibly having to kind of build these blocks on top of them that people can then compose. And it sounds like depending on the level of skill of the end user, they'll be either using the most based primitives or some of these building blocks that you're building.
Can Duruk [00:35:13]:
It's not even just skill. It's just you know I want Felt to get out of people's way. I don't want people to learn this FSL, like learn this spec. they should be able to do a simpler thing. They want, I mean, I think that's like one of you guys, like core value propositions, right? You don't want people to worry about billing. There's probably ways to hook into the deeper of Orb but most of the time you just want to be like, give me a subscription, whatever.
Kshitij Grover [00:35:40]:
Yeah. Exactly. Exactly. Yeah. Well, no. This has been really fun. I think maybe the last question is, what is something that's coming up? It could be any of the things you already described that you're most excited about. It could be a technical challenge or a product feature, but they're just curious personally, what's the thing you're most looking forward to?
Can Duruk [00:36:00]:
I mean, it's just hard to pick, and I know this is a cagey answer. I think live data is gonna open up a lot of different use cases for Felt. I think that like close out the suite, like the Felt suite. After that, there's gonna be a sales and marketing machine. Like once we have live data, that's gonna be the big thing. And on top of it, I mean I'm just, I guess like live data and API use cases. It's just what I'm most excited about. I think we've solved a lot of the hard engineering problems. Now we need to build the plumbing around them.
Can Duruk [00:36:38]:
So now people can actually like to get a lot of the power that Felt has for itself.
Kshitij Grover [00:36:43]:
Yeah.
Can Duruk [00:36:43]:
That's really a pattern that we've been following. Our upload tool used to be internal only. We started with this idea of that. We had this data library. We're like to build this like the most curated beautiful data library. Turns out people want to like having that own because it was way too powerful. People were I would have my own data on Kafka. They were no, no, no, you don't.
Can Duruk [00:37:04]:
We got this. But it turns out people might have their own data. So I think opening up more of Felt's powers, like more people programmatically. Or like, you know, programmatically just via APIs or connections. That's what I'm most excited about.
Kshitij Grover [00:37:19]:
That makes a lot of sense. And, excited for you to build that and excited to see more Felt maps out in the wild. Thanks for the conversation, and thank you for coming on.
Can Duruk [00:37:28]:
Yeah. Thanks for having me.