#352 How Unstructured Data is Shaping the Future of AI with Kirk Marple

00:00:00]

Mehmet: Hello and welcome back to a new episode of the CTO show with Mehmet. Today I'm very pleased joining me Kirk Marple. Kirk is the founder and CEO of, uh, GRAPHLIT I hope I pronounced it right, Kirk. It's GRAPHLIT. [00:01:00] Yeah, GRAPHLIT sorry, um, and you know, the way I love to do it, as I was telling you, Kirk, is I keep it to my guests to introduce themselves.

Mehmet: So tell us a little bit more about you, your journey, and you know, what led you to found your company. So the floor is yours.

Kirk: Yeah, thank you. No, great to be here. And, um, I mean, I'm a career software developer. I always say, I mean, I've been doing it for gosh, I mean, 30 years now and, uh, was at Microsoft in the early days, um, basically had been there after my master's and then founded a company where I had for about 10 years.

Kirk: And then, um, sold that company and was CTO of a couple of different places, um, in the unstructured data space. And then founded this company about three years ago.

Mehmet: That's cool, uh, Kirk. And, uh, thank you again for being here. Now, usually it's kind of became my traditional question that I ask every time I interview founders like yourself, Kirk is, you know, what was the moment that you saw something missing in the [00:02:00] market?

Mehmet: So you decided to start Graphlit, like, you know, and knowing that unstructured data as a, you know, Uh, important topic, and it's something that I like to discuss it because little bit biased working in that field for quite some time. So, So share with us, you know, what was the, the vision that, uh, you know, you started with and the gaps in the market that you saw that, you know, decided to start Graphlit.

Kirk: Yeah. I mean, it really goes back to the days of the, um, broadcast video transcoding company that I had. And we really talked about back then it was always called file based workflows. So there were these stages of. Processing of the files, I mean, ingestion, metadata, indexing, processing, and then, uh, delivery or publishing.

Kirk: And when, after I sold the company, as I said, and I was working in the, um, the autonomous vehicle space, uh, [00:03:00] general motors and helping them build an unstructured data pipeline to take, um, data that was not just video, but it was like video telemetry, LIDAR. Um, and get it in the hands of their data scientists.

Kirk: And so it kind of pulled from my background in the video processing. And my biggest, I mean, initial thing was, I mean, the tooling doesn't exist at the time, and this was six, seven years ago or so, and there just isn't anything off the shelf, not like a, like a spark or I mean, a five trend or snowflake, these things just didn't exist.

Kirk: And so I just started, I mean, kind of thinking about it back then of like, Oh, okay, if I want to start something new. This is an area that that's untapped and the couple of companies I worked at after that were pretty similar I mean building on structured data pipelines a lot of computer vision a lot of um, I mean really pulling together different media types and Um, basically three years ago when I got started I looked at it and said look, I mean It's only going to grow and the tooling's not [00:04:00] there and i'd love to be the one to build

Mehmet: Great insights kirk and you know especially when we talk about data in general, but let's focus a bit about unstructured data.

Mehmet: And, you know, I always tell people like unstructured data, it's very unique because it has a lot of insight that usually, you know, organizations, they are not aware of. So in your opinion, what makes unstructured data really, you know, unique compared to structured data and why it's important for businesses to really manage it effectively?

Kirk: Yeah, I mean, I've, I've spent most of my whole career dealing with, I mean, audio and video and images. And I think a lot of people, I mean, it's the classic, like, is it really structured? Is there, is there structure there? And I mean, what I learned from, gosh, I mean, almost my first job is metadata. And it's so, there's so much to it of it.

Kirk: I mean, there's bytes of a file, like a TIFF file or a JPEG file. But there's always [00:05:00] typically a metadata header in there. So, I mean, you figure out, I mean, who's the author of it? I mean, what camera was something taken on? What the GPS locations, um, are? And I always found that really interesting because, I mean, it's, it's kind of, um, another, I mean, it's not the typical rows and columns kind of data that the structured, I mean, SQL data is, but it's, it really varies.

Kirk: And there's so many different formats and so many different flavors. Um, and it's, it gets super complex, but I mean, it's, it's just. Randomly, I kind of spent my whole career in this space that dealing with so many different file formats and building file format parsers, um, it's, it's really a kind of all drawn to this point where now we can, we can support all these different file formats.

Mehmet: Absolutely. And, um, you mentioned something which is crucial is the amount of metadata that exists within the unstructured data. Yeah. And I know like, uh, you know, Here, I want you to a little bit explain to us like also about like the knowledge graphs [00:06:00] and how these, you know, can enhance the capabilities, of course, when it comes also to AI applications, uh, and maybe building workflows, um, when we think about unstructured data.

Mehmet: So if you can, I know it's a bit technical, but it's very important, I believe. So we can understand, you know, the, um. The benefits out of that.

Kirk: Yeah. I mean, what I, what I found is, I mean, especially in the imagery and video space when I was working in the drone area is there, like say a DJI drone actually captures a ton of metadata about what it's doing.

Kirk: I mean, the altitude, the speed, the lens, if it's doing thermal data. Um, and it's really interesting that there's a, there's a huge payload of data. And I sort of got this idea when I was at General Motors of, Kind of index everything in time and space and so we create packets of information of I mean an image that existed of looking at something at a specific point in time and was in a specific like geo location [00:07:00] and I started to get this concept of the interrelationships of all this data and like if you're doing LIDAR.

Kirk: 3D data. It's, I mean, another representation of the real world that kind of overlaps and can get kind of organized with an audio recording at the same time and place and an image at the same time and place. And that's where the metadata really comes in of it provides you a, I mean, in SQL terms, it's kind of like a key of here's something I can index on.

Kirk: And it's not just I mean, a number or a string, it's, I mean, a geolocation or it's a time range. Um, and that really gives this kind of multi modality to the data. Um, and I think that's what's super interesting about this, where it's, it's just, it's a lot more variance. And I can roll into, I mean, the knowledge graph side of this is really all about those interrelationships.

Kirk: And I think, you know, Every node in the graph for us, at least it's a piece of content. So it's an audio video [00:08:00] document, a webpage, um, but it's also what we call observations. And so it's things that are observed in the data, which could be, I mean, like in, This it's like, uh, you and I are, are persons that, I mean, we're guests are on this, on this podcast where the podcast is a piece of content.

Kirk: We are person entities and you'd have a place entity because I'm in Seattle right now that would have a relationship. And if you think about the depth of all this data that is sort of ambient, um, that then becomes your knowledge graph. And then the cool part about it is you can query on it. Um, so you could go back and be like, Hey, find me all the podcasts that I recorded with Kirk when he was in Seattle.

Kirk: And essentially you now have an index. Um, and that's, that's really where the, I mean, the knowledge graph starts.

Mehmet: If I want to compare this to the way an Excuse my ignorance here, Kurt. Like, if I want to compare this to the [00:09:00] way how, for example, large language models work, for example, like, because what I'm trying to, to, to make an analogy because you just mentioned something like, for example, if I, now in the language of, in the, you know, sense of large language models.

Mehmet: So basically I give knowledge and then, you know, the model. is able to, you know, get some abstract from whatever knowledge we have provided the same way you said, like, for example, I asked, so how, how this is like similar or different, I would say from, from that perspective.

Kirk: Um, I mean, it's interesting. I mean, I'm, I don't profess to be a data scientist or kind of an expert in the machine learning space.

Kirk: I'm more of a application developer on top of the models, but I mean, I think there's, there's a data representation. That is somewhat like video encoding or image encoding where you're compressing, uh, I mean, uh, sort of a high dimension of data down to a smaller dimension. And I [00:10:00] think, I mean, from the way I understand is, I mean, the neural nets, I mean, they're keeping relationships between, um, the data, the tokens.

Kirk: And I think it's, it's a, maybe it's a more of a hidden relationship that it's not obvious, like to, to V you can't really visualize it the same way where a graph it's nodes and edges, and it's a bit easier to visualize. Um, I think there's definitely a relationship there. I mean, there's a relationship in how like vector embeddings and different ways of, of kind of querying that data, um, are stored almost in a, in kind of a graph.

Kirk: Um, kind of model, um, for retrieval, but I think it's, I mean, it's definitely, there's an analogy there, but I'm probably not the right one to answer the, the technical details on the, on that side of it.

Mehmet: Absolutely. Just, it was, you know, something that popped in my head, uh, because you just gave the example, which is great, by the way.

Mehmet: Now, if, um, you know, I'm not mistaken, also Kirk, like You do a lot of automations when it comes to the workflow. So if you can [00:11:00] explain to us, like, you know, how, because I'm, I'm, I'm a big fan of anything automation regardless. So I'm interested to know, you know, what are exactly, you know, the use cases is when it comes to automation and how this is, is measured from, from, you know, whoever is using a Graphlit, like from time perspective, is it like, effort that they need to put.

Mehmet: So if you can go over this, uh, to us, Kirk.

Kirk: Yeah, for sure. I mean, I've always been a fan of kind of formalizing, um, what we call workflow stages. And I think people might be familiar with, I mean, DAGs or directed acyclic graph for data engineering, where you can kind of say, I mean, your pull data from here, Do a transformation on it, load it into a database.

Kirk: And what we see, and this is something I kind of, um, rift off of in my old company as well, but we start with ingestion and that's kind of like, okay, where do you get the data from? Are you downloading it from the web? Are you pulling it from Google drive? Are you grabbing it off blob storage? And so we go ingestion, then we go into [00:12:00] indexing and that's okay.

Kirk: I have a file or have, um, a URL to something and I'm going to get metadata out of it. And so indexing is that process of going kind of content to metadata. And then what we do is, um, we can have extraction, which is like entity extraction, like pull people, places, and things from the data. And, um, Oh, sorry, before that is preparation.

Kirk: Preparation is kind of the idea of, um, extracting text. And so that could be, I mean, PDF extraction, audio transcription. Um, so what, after you index it, we get the, we prepare it and then we can extract it to a knowledge graph. Um, we can then enrich it. So enrichment is the next stage. And these are linear. I mean, you can, we go in sequence.

Kirk: Um, and then enrichment is typically like, okay, I've grabbed a person. Um, all I know is its name, uh, the person's name, um, I wanna go query another database, or I want to go query Wikipedia and get some more [00:13:00] metadata around that. And that's what we call enrichment. Um, that could just be an API call to a CRM or call it Google Maps, um, because all we have is a place name.

Kirk: Um, so you're kind of adding fidelity around the data. And then at that point, I mean, we're basically, we've, we've fully ingested the content and we would consider the workflow pretty much complete. And then it goes into a consumption of, do you want to query it? Do you want to have a conversation with it, with an LLM?

Kirk: Um, or do you want to re like what we call publishing, kind of repurpose that data. And so. I've always described this as like a two sided funnel that the first step is kind of the funnel of the data in to the system, and then the next is kind of the funnel going outwards and that that's really how we model all of our workflows.

Kirk: And then we decided to be, I mean, I've always kind of been prescriptive about that, but allowing configuration. So at each stage of the workflow, You can configure different steps like what [00:14:00] transcription model do you want to use? What lm do you want to use? Um, like embeddings model do you want to use stuff like that?

Kirk: And so we do let you Configure different knobs, but it's all kind of within the same flow

Mehmet: I got it now i'm big fan also kurt in getting The technology out and trying to convert it or translate it into business benefits So let me start asking you like where have you seen? You know, the best use cases from verticals perspective, like, is there any specific vertical that you found that what you do really, you know, it met exactly what these guys were looking for?

Kirk: I mean, it's, it's interesting. I mean, it's, we're, we're really horizontal. So we're not, we try not to be too specific about a vertical, but what we've seen, especially in like healthcare, um, there's some interesting cases with, um, even just the marketing groups within healthcare companies or pharmaceutical companies, they have a lot of technical [00:15:00] material.

Kirk: Um, they're trying to write essentially blog posts, or they're trying to repurpose content, they're doing social media, um, and having the ability to have an LLM sort of repurpose, um, and essentially translate into, um, a more, a friendlier tone, a friendlier, um, way to read, um, is super useful. But I mean, the hard part for them is, I mean, that data might exist on SharePoint.

Kirk: It might be in a PowerPoint. It might be in a PDF. So all of the stages of the workflow we talked about are really valuable just to get the data. to have some velocity to it, like to get the data moving in a way that then they can consume it. Um, and so it's not necessarily specific to healthcare, but I've, I've, for some reason, I mean, we've talked to several, um, folks and it's, it's very valuable for, um, especially the knowledge graph side, because they have a lot of different terminology, a lot of different, um, entities they want to track.

Kirk: And, um, and even recently with, um, Uh, the concept of graph rag, um, kind of integrating graphs with, uh, retrieval [00:16:00] augmented generation, um, has been very appealing to healthcare. Um, but we've seen FinTech. We've seen kind of more entertainment. I mean, or, or sorry, consumer focused, um, products as well. Um, but it's, it's really anybody that has domain specific data we talked about.

Mehmet: Yeah, healthcare makes a lot of sense, uh, Kirk, because, you know, when I interviewed some, um, data scientists in the health care or even like healthcare professionals themselves And you know the common theme that always used to come they have a lot of actually data points Yeah, and it's very very unstructured by nature, you know Whatever.

Mehmet: It's like the, uh, you know, when, when you do city scan or when you do like an x ray, like plenty of things. And of course the prescriptions that, you know, like sometimes the doctors will be writing, whatever the nurse will be writing. So they have plenty of data points over there. You mentioned something interesting about drag, which is retrieval augmented generation.

Mehmet: So You know, just for the sake of [00:17:00] people who didn't hear about the term. So this is now the common term we use about how we can prevent AI from hallucinating. So I'm interested to know, uh, from your perspective, Kirk, like, uh, and like how the rag works, you know, in simple terms, in layman terms, and What could be the benefits in the context of the unstructured data management?

Kirk: Yeah, for sure. I mean, it's, it's interesting because it kind of parallels how, uh, really my company started is. I mean, we started on the data ingestion side and then the retrieval sides. We were building, I mean, a search and visualization for unstructured data. Um, for companies in the built world, like construction companies, ports, um, like railways, and they had a lot of data, um, say like inspection reports or audio transcriptions that they'd want to have from like, um, when they do like a field field inspection.

Kirk: And so we were kind of working, I mean, As more of a search problem of, okay, we got to get data in [00:18:00] here so we can search, they can visualize it. They can actually track all their data. And what this term rag started, I don't know, what is it? 18, 24 months ago now, um, kind of being the pattern of how do you feed Domain specific data into LLMs, because as I mean, people probably know is the LLMs are essentially read only.

Kirk: I mean, they're trained on a data set and you have to basically give them data that is private or kind of personal for your situation. And RAG is essentially that pattern of doing a query. Typically using a vector database to kind of find similar data, formatting it in a way that is useful for the LLM, which there's a lot of secret sauce there, which we'd go into.

Kirk: And then basically almost like you were just in a chat GPT. I mean, you can essentially paste in the prompt that you would hand, um, from rag to an LLM get and get the data back. And that would be your, your response. Um, but then there's, there's a lot of other [00:19:00] kind of secret sauce there around, um, yeah.

Kirk: Making sure it giving it's going to give you JSON output so you can get some structure and know that it gave you a quality response. It didn't get truncated and then just the quality of what data you're giving it. The order the rank of data format of that data. There's a lot of details that go into making a good rag solution.

Mehmet: Absolutely. And thank you for sharing this Kurt. Now, talking about AI, which is like very, um, you know, exciting topic and with everything that's happening, uh, you know, these days, um, are there like any specific trends within the AI that can be, I would say, coupled with unstructured data, uh, That get you excited about and you think like, okay, like this gonna really shape the whole industry.

Mehmet: There are any specific things on the radar. I would say, [00:20:00]

Kirk: I mean, it's interesting. It's I mean, every week it seems like there's something new and I think there's I mean, just the text side of it, documents, web pages, um, I mean, it's starting to standardize, I mean, I think, um, but it's, there's still a need for like AI driven web scrapers.

Kirk: I mean, to get the right data out of web pages, really good PDF extraction. I mean, there's whole companies that are spun up just essentially to do, um, do that these days. But then, I mean, one of the things I saw probably in the last 12, 18 months was the cost and quality of audio transcription went way, way down.

Kirk: And it was really kind of a turning point for a lot of applications that really had potential, but now became, I mean, really commercially vibe. And I mean, we're using a company called Deep Gram. Um, there's other companies like, uh, assembly ai, um, OpenAI has their whispers algorithm and. It's, I mean, it's become just so dead cheap now to, I mean, transcribe a podcast or transcribe meeting meetings.[00:21:00]

Kirk: Um, so I think that that's another area that that's been valuable. And I think video analytics, I mean, video analysis is just, I mean, we were doing video analytics, what, five, eight years ago, uh, But I think now with the new multimodal models, there's so much more capabilities. Um, and that actually is something I did a project on of just taking, um, basically inspection images like of, um, for maybe like a, a virtual kind of insurance use case.

Kirk: And I, I fed it a bunch of images and, um, to a multimodal model and asked it to like, um, it was a fire that was, and there were fire trucks and asked it, I mean, sort of, Um, give me a score, a rating of how, how serious is this and, um, what else can I, can you take away from that? It's like doing an inspection report and it's incredibly good.

Kirk: I mean, it does OCR to pick up, like, it's the Chicago fire department that was there. It could tell if there was smoke. It could tell the fire was still going on on the, on the roof of the building. [00:22:00] And I mean that getting back to automation, there's just so much capability now. To tie together all these modalities of it's not just a computer vision problem.

Kirk: It's not just an NLP problem. It's really a multimodal solution. And that's what really excites me about pulling all these different pieces together.

Mehmet: Yeah, I think we just start scratch the surface there. And, you know, I'm not sure. And by the way, like, I remember when I started, you know, Tragedy was still new when I started the podcast.

Mehmet: So I was asking the question, okay, what do you think the trend is in five, six years? And I stopped asking this question because no one can, can, can guess what would happen in one week time, I say. But of course, we can spot some trends. And this is why I asked you about, you know, and what you mentioned is also, I started to see a lot of people building on this.

Mehmet: I mean, uh, transcribing audio. And not only necessarily podcasts, like podcasts, meetings, and, you know, they're building a bunch of applications on top of that, which is really saves a [00:23:00] lot of time, you know, and it's like a dream for having such technology. I want like a little bit to shift the conversation now, Kirk.

Mehmet: And, you know, I want to ask you. Because you've done this and you've done it very well. So when you are bringing new products to the market from the ground up, um, you mentioned the word secret sauce and I know like there's no one secret sauce, but I mean, what worked for you in the process? Was it like hiring the good team?

Mehmet: Like, was it implementing the right technology while bringing this product? You know, what can you share with us about your experience?

Kirk: No, I think it's, it's such a great question because as a technologist, I mean, I'm really, I mean, more on the, on the technical founder side. And I think, I mean, the biggest learning is, I mean, for anybody kind of bringing a product to market or a company to market is, I mean, figure out marketing first.

Kirk: It's just, it's such a hard thing to figure out awareness, like building the product. [00:24:00] I mean, once you've been like, can find a good team, have a good idea, the technology becomes so easy, but distribution of the product is, I mean, it's really a lot of the stuff that. It, we, I mean, it's, it's a struggle. I mean, how to getting awareness on social media, getting, I mean, how do ads work and kind of, I almost have to do like a whole master's degree in, in all this area, in addition to kind of the technology side.

Kirk: And so I think anybody bringing a product to market is, and it's, it's such almost a meme these days of like only second time founders figure out how to market, but it's, it's kind of true. I mean, it's, there's so much to learn in that space, but I think the key thing is having a vision. And having the grit to kind of follow through with it.

Kirk: And I've pretty much had the same vision for what this product could do for six or seven years. I mean, I was thinking about it as a podcast discovery platform actually. And it was before transcription got really cheap, but I had this idea of [00:25:00] there's so much information embedded in discussions and in, um, kind of different multimedia material and the ability from a learning platform.

Kirk: Um, not just academic, but in business and personal to explore that data. And that's really what our kind of sort of vision is and Northstar is having a platform that lets people build knowledge driven applications where we make the ingestion easy and you just have to figure out how you want to consume that data, um, and hopefully focus on the UX and stuff like that.

Kirk: So I think it's, I mean, there's so many pieces to it, but I think you, you have to really, I mean, you have to have that vision that you can keep, um, sort of through the good times and the bad times of just keep focused and just really never giving up is a great solution. I mean, it's a great way to kind of get through it because, I mean, we had to, we went through ups and downs and we had to pivot, had to do some layoffs, had to come out the other side, refocus on [00:26:00] kind of where we're, where we're heading because you can't control the market, you can't control the economy.

Kirk: And, um, but, but your vision can, um, Cannot quit. I mean, if you don't want it to, so

Mehmet: absolutely. And I'm happy you brought like, um, you know, the concept of,

Kirk: you know,

Mehmet: it's not only the second time founders because some people, you know, they think, Oh, you know, they have done it multiple times. So it's easy for them now.

Mehmet: But I always remind people, okay, the second time or third time founder one day he was a first time founder as well. So yeah. If he or she, they succeed. Okay. It's not easy to your point. Um, but yeah, so, so, and, you know, mentioned something very crucial. I, I believe Kirk, which is the, The vision and you know, what are you trying to achieve with what you are building?

Mehmet: This is more important and I believe you know, like once you start to build and talk to people Because I come from technical background and then I shifted into more like sales oriented roles in my career So yeah, but then you can start to relate. Actually, I [00:27:00] was selling somehow, you know, because when I was talking about my idea and trying to convince people to this kind of a sales, right.

Mehmet: So, and, you know, and started, you know, as technologists, especially here where, you know, I grew, like, it was not like very bright, uh, image for someone who sells. So, so you need to, I need to brainwash my, my own mind say, okay, no, like you're actually doing something good. So absolutely on this point was occurred.

Mehmet: Um, now shifting from, from, from being, you know, you work as a CTO and then you became head founder CEO. But I can see like you still are. technical, um, part of it, which is great by the way, but I'm sure like you have, you know, a way to balance between the technical and business aspects in, in, in, in, in a tech startup.

Mehmet: So what, like some of the key lessons that you have learned that you can share with us today?

Kirk: Yeah, it's a, it's [00:28:00] definitely a balance because. I mean, I, I built the backend platform, but there's so much more than that to, to a product. I mean, there's, I mean, like the UI for a developer portal, um, like whatever user interface you have, there's the developer relations side of it.

Kirk: There's, I mean, the sales and marketing and social. kind of media side of things of just getting the, um, getting it all out there. And there's, there's the operational side, the finance side. Um, so I think it's important to surround yourself with people that compliment. Um, I mean, I think for, for us, it's still, I mean, I'm not giving up the development, um, for the platform, but I mean, I need to have a good UI dev.

Kirk: I need to have a good designer, um, people to help with marketing, a good finance person. And I think there's, it's, I kind of look at it where. You kind of have to work in chunks. Like a lot of times the early part of my day is meetings and kind of more my business hat and then the end of my day and night are coding.

Kirk: And I've kind of always been that way where, I mean, almost [00:29:00] until one or two in the afternoon, I'm just kind of on business calls or doing podcasts or sales calls or whatever it is. And, and I think it's, it's hard to, I mean, context switch. And I think that's the biggest learning I've had is stay focused, Get like chunk your time up.

Kirk: I mean, maybe three hours at a time on, on one sort of theme of the business. Um, cause it's really hard to flip back and forth between like, you're debugging something and writing some code, and then you're going over and doing a sales call and then having to jump back into the code and, and that, that just doesn't work.

Kirk: So I think kind of chunking up your time and staying focused on tasks is, is probably the hardest and most important thing to do.

Mehmet: Yeah, you know, makes sense. Makes sense to me. Uh, at some stage, you know, when I shifted from kind of a pure technical to a hybrid role, you know, it wasn't the best thing to do because you live two roles in the same time.

Mehmet: And yeah, as you mentioned, you know, the best way is to chunk the time and try [00:30:00] to, you know, and when something that worked for me is when I'm wearing a certain hat. I'm wearing that hat. I'm forgetting about the other hat because you know, people I need to give the people, you know, the, the, the, you know, uh, let them feel that.

Mehmet: Okay. The person if I'm technical, let's say, okay, this guy is a technical guy. He's not like a, you know, just a marketer or like a sales guy and vice versa also as well. So I needed also to learn how to adapt this. Um, occurred like really, you know, the, the journey is, is, is, is so, um, rich, I would say. So, how hard do you think today is It's for someone to start a tech startup if they come from a technical background, what can you share with us?

Mehmet: I'm trying to ask you this question to inspire more people to start actually,

Kirk: I mean, I think it's never been easier. I mean, really, given there's so many parts to that, that answer. But I mean, part is just the technology. Um, it's, I mean, there's [00:31:00] tech platforms, I mean, cloud platforms, web frameworks, I mean, all the pieces that you need.

Kirk: Um, really anybody, I mean, can, can start and it's very, very cheap. I mean, a couple of bucks for, for sell every month to just have a web, I mean, web hosting, or you could do it yourself. I mean, even now there's, there's just such easy, um, easy approaches for that. So I don't think, I don't think getting something built is very hard and pretty much anybody with any tech skills can, can do it.

Kirk: Um, I think the harder part is, I mean, Number one, I mean, kind of developing an audience for distribution and figuring out, okay, I mean, if I build it, how do I get it out there? Because it's, it's such a, um, what do you want to say? A noisy environment that if you build it, nobody's going to know about it unless you market it.

Kirk: And I had been in a different space where the broadcast video space. We would go to two trade shows a year and see all of our customers. And it's a completely different sales environment that, I mean, we didn't really have [00:32:00] to market per se. We could just, I mean, we would, it was more direct person to person selling and relationships.

Kirk: And so nowadays with, I mean, being in the developer tool space or kind of cloud services space, there's just a billion companies popping up and sort of separating, like, how do you separate yourself and get awareness is, is I mean, it's honestly, I don't know, for me, I feel it's almost 70 percent of the problem these days.

Kirk: And so for a new startup coming in, I think what I've seen, the good successful ones is they develop like a social media following. They have a YouTube channel or they have a good Twitter following and that. Drags them along really far without, I mean, even having maybe a great product. And so I think, um, anybody starting, I mean, don't, don't forget about that part of it because it's, it, it can be almost more valuable than, than the technology, um, I think for harder tech companies, like, I mean, like ours, where we're essentially building an, a software infrastructure, um, that is much more like a [00:33:00] snowflake or, I mean, like an Azure type service that a cloud service and it's, um, It takes a while to bake in the oven.

Kirk: And so I think those are different kinds of companies where I think what we've found is really having a, I mean, we're selling to developers. You have to think like developer. What do people like? They like good docs and they like sample code and they want to be able to like try it for themselves, self service.

Kirk: And so I think you have to kind of know what vertical market you're targeting and do the right things to, make those people happy and, um, and, uh, reduce friction.

Mehmet: Absolutely. And, you know, I can add to what you mentioned Kirk about, you know, building the audience and, you know, again, like many founders, they were mentioning the same thing that nowadays buyers, whoever these buyers are, like whether they are like deep technical people, they are business people, actually they know I mean, they know that they need a solution, right?

Mehmet: So you need to [00:34:00] be present in the market so they can find you out and this is where you differentiate yourself and One of the things, especially like product like yours, Kirk, what I've seen, like if you build a community of people who, you know, I can say like, yeah, hardcore tech people, let's say, like who likes the tech and you know, like you, you build this community and then they will be, they become by the way, your marketing engine somehow, of course, you don't rely 100%.

Mehmet: And, you know, I, I try to explain to people like to your point, whether whatever product is FinTech, Um, even not related to technology, you cannot like expect that I will throw this product in the market. I mean, like I will put it and then people will come to you unless you, you make some noise, right? So you need to make some noise.

Mehmet: So people will get to know about you and say very noisy world, as you said, like it's plenty of, of, of noise. You know, tech companies that starts here and there. So absolutely. To your point, it's very, very, very much value. Yeah,

Kirk: open source is another another good point. We see a [00:35:00] lot of companies these days that are kind of open source as a marketing effort where they just start with an open source project and they try to figure out monetization later.

Kirk: And I think it's it's I mean, it works for the community part, but it doesn't always work for the monetization part. And I think it becomes a difficult thing where like, I mean, okay, we've now have all these users and people wanting, having it for free, how do they turn the pay switch on? And I think, I think that's a, you got to think about that when you start a company, because it's easy to kind of get locked in where if you have.

Kirk: users that aren't really ever willing to pay. Maybe they're just hobbyists or college students or whatever. Um, it may be great to get a lot of GitHub stars, but you're not going to make a company out of it. And so. I mean, and we're kind of doing the opposite where it's like, look, we're closed source, but with open source SDKs and open source layers on top of us, but we're just an API and a pay as you go is API, which is you have to do.

Kirk: I mean, we're, [00:36:00] we're free to use, so that's, I mean, how do you get people in, but you don't get the community necessarily day one because you're, you don't have that open source part. Um, but it's but we're trying to build a business, and so it's there's kind of two sides in that coin that you have to find kind of a spot, a good spot between.

Mehmet: Absolutely. And you know, I've, I've checked, you know, before your, uh, your website and I've seen you have like kind of a freemium, which is, I'm big believer in it because this is get traction, let people have a, uh, I would say like sense of how the products work or how the service works. So absolutely on this.

Mehmet: Uh, Kurt, like we, we covered a lot today and you shared a lot of valuable knowledge. Uh, final, you know, words of wisdom, I like to call them and where people can find more about Graphlit and more about you.

Kirk: Yeah, no, thanks. Um, let's see. So we're on Twitter, um, at, at just, uh, Graphlit and myself just at, at Kirk Marple and, um, we have our, our website Graphlit.

Kirk: com and [00:37:00] as you said, I mean, we're free to get started with. We have actually a pretty healthy. Free tier that people are building. I mean, um, hobbyist applications on and love to have more developers. Try us out. We just released a Python and a TypeScript SDK. Um, but yeah, happy to get in touch with anybody that's kind of wanting to build apps in the AI space that can take advantage of it.

Mehmet: Great. Thank you very much, Kirk, again, for the time. And I know it was a little bit late for you, uh, in the night. So thank you very much for, for being here today and sharing, uh, your knowledge and sharing also about what you're building. And this is for the audience. Uh, this is how I end my episodes. If you just discovered this podcast, by luck, thank you for passing by.

Mehmet: If you liked it, please subscribe, we are available on all podcasting platforms and we are available on YouTube also as well. So please subscribe, share it with your friends and colleagues. And if you are one of the. People who keep coming back, the loyal followers who keep sending me suggestions, comments, and you know, all their notes.

Mehmet: Thank you for doing so. Please keep doing that. And thank you for all your support. [00:38:00] Uh, I hope you enjoyed this episode. Thank you very much. And we will be again very soon. Thank you. Bye bye.