 Telmay provides a full end-to-end solution around data reliability, if you may think about it, with the foundation of data observability. Hi, this is your host, Swapna Bhartya and today we have with us Mona Rakhibe, co-founder and CEO of Telmay. Mona is great to have you on the show. Good to be on your show, Swapna. And it's my pleasure to host you today. Since you are also co-founder and CEO of the company, I would love to know a bit about the original company, what led you to the creation of the company, what problem do you folks solve for the larger ecosystem and also what is the story of the name? That's an interesting one and it will be weaved into the story. So Telmay is a data observability company. So what that means is we look at the data from ingestion to consumption of the data to predict the health of the data. And we started Telmay a little over two and a half years back. Me, I'm the co-founder and CEO of Telmay and my other partner, Max Lukachev, he's the co-founder of CTO. We've spent a lot of our career in enterprise data ecosystem and what we realize in just the last five to eight years, companies have shifted a lot around their data strategy like every company is talking about real-time analytics, ML AI based initiatives, the volume velocity, the scale at which we are operating data has shifted tremendously, but what did not change about the data ecosystem is how we looked at data and how we felt confident about the health of the data, the reliability of the data, did we trust the data. Teams were still using old school tools and technology which were built on legacy system to predict the health of the data, validate this. And Max, who's my co-founder, worked at SignalFX, which is an infrastructure observability company. I'm a product manager, engineer, turn product manager, turn co-founder. Max has always stick to the line of technology and data science. And when he worked at SignalFX, he felt that the technology and the scale that we have today, we can solve data reliability using AI techniques, statistical analysis in a nutshell, high compute, how can we use low-cost compute to calculate and figure out the health of the data? So that really gave birth to TellMy, TellMeAI. That's how our name is, TellMy. So it is the sigma in our name is the mathematical foundation of everything we do. It's the summation. And till date, it is the strongest foundation that we have that every problem we look at. And is this a problem that needs scale? And if it needs scale, it needs to be done in a mathematical manner. So that's a little bit about the name and the problem we are trying to solve. The company, as I mentioned, is almost three years old. We are a seed stage company, eight and a half million raised from Glasswing Ventures, Zeta Venture Partners, Fortosex. We were also in Y Combinator summer 21 program. We built partnership with Google, Databricks, Snowflake, and have customers like Merkel, ClearBit, Datastax using the product already. So that's the long-term shot of me, the company and the name. If you look at the observability space, it has evolved over time from the days of logging, tracing, monitoring. The scope has also grown a bit, but it has also become a very crowded space. There are a lot of companies who are working in this observability space. Talk a bit about how you have seen the evolution of observability and what specific niche that you are kind of addressing in this crowded space where you are not just another player, but you found some gaps and you felt that, hey, you talked about signal effects that, hey, these were the niche that need to be fulfilled. These are the customers who needed help. That's where Telma is coming to help them. So two, four question, evolution and your role. The evolution has been like, if you look at observability, it's been existing from control theory times, right? Like, how can you look at the external systems features and then predict the health of any system, right? So that's like old definition of observability. In the recent past, the tech world embraced that philosophy in many areas from API observability, infrastructure observability. So the companies like Datadog, SignalFX and others started using those philosophies to predict the health of infrastructure. Now, in similar manner, so in my opinion, observability is a concept and the use cases are evolving and the different industries are adopting it for different use cases. So with that said, the first thing Telma is specifically in data observability space. That means we look at the data itself, like the data that's fueling your key business decision, the data that's fueling your ML and AI initiatives, and we look at that data and observe and predict the health of the data. So within the data observability ecosystem also, like gigawomb data report, which we shared with you as well, and there were 15 top vendors. So although the category is four or five years old, there are 15 top vendors. So talk alone about the top vendors, then imagine how many vendors are there. The problem is very widespread. So obviously there is a need in the market, so companies are identifying that needs and then there are many companies evolving in that space because of the need. Now, how do we distinguish ourselves in this hyper-crowded mode, right? So to be honest, a lot of people ask me, really, you want to do data observability? It's already like a Red Sea. So many companies are already there. What Matt sent me really had a very, very strong intuition was how we solved this problem. First thing we knew is data quality and data validation has to be shifted to the left of your data pipelines. That means where the problems occurred to the source system, which meant that we had to have a system that supported any kind of underlying sources, right? Whether it's a legacy system on-prem data storage, even streaming like Kafka, PubSub, or Red Panda, for that matter, or the modern data warehouses, cloud-based data warehouses, Delta Lake ecosystem. So having the full breadth of the ecosystem needed a different architecture to kind of read data from any system and process it in a decoupled manner. So that is our number one differentiation and how we do it and how we support different data sources in different format and so on and so forth. The next thing is we also knew true data quality can be achieved only when we find accuracy and where the data problems are at column value level, right? So we go super deep, do machine learning based anomaly detection. We can find things like value outliers, out-of-range values, statistical drift, over time, how this data has changed. So we go broad, but we also go deep. And the last thing is how do we do it without adding any latency on your data? Most people want the results and the data pipeline to be flowing constantly without any interruptions, without any excessive cost. So we've also made sure that the way we have designed Tellmai is that we keep all of this with keeping the cost low and the latency on your data pipeline. So those are three fundamental differences which we designed Tellmai with and that's starting to give dividends as we are accelerating in the market. And as you know, we were like a leader in Gigaum report in spite of the fact we are relatively younger, much less funded, but this message around how we have designed it and how we saw has received a very good validation from the market. There needs specific industries of verticals that you folks are serving, or it doesn't really matter whichever, whoever, to this day's cloud native, you have to have observatory strategy in place. But I would love to hear about some use cases or the industry that you can do. So the solution is very agnostic to vertical in general and we try to do a lot of research of if there are any industries that would be early adopters, so this is a new category, right? So there's not an existing budget, there's a lot of awareness we have to do. We noticed that there are certain scale of companies and industries that understand specifically if we do it with the data quality and reliability angle. So there are verticals like D2V data vendors or like financial industries, healthcare, those companies are the faster ones to identify this problem and understand the solution differentiation. We as a company have not done that segmentation heavily yet. We are getting like opportunities from all over the case and the platform is architected to do, support these verticals without any adjustments, right? So in the end, it's a data warehouse, it has data which is all over the place, different type of volume. However, what I noticed is that industries which have reached a certain level of maturity with their data initiatives have identified this problem better and most of the times these are enterprises. So mid markets or smaller companies are still or digitally native build companies are still not hit the problem of scale and solving. So the tendency of people is usually let me build some validations and we are selling to highly technical people, right? Like these people can build anything, right? So there are always this natural inclination that let's start building something. And then they hit the problem of like, should we sit and maintain a tool which is for data observability, data validation or should we build what is core to our company? And that realization takes place faster for larger companies who have hit that maturity on stage. Also an ML based anomaly detection takes much more investment to start building to maintain and so on. So it's a little bit more on the phase of the company, the realization of the problem than the vertical itself. These days, one of the, you know, hottest technologies topic is generative AI. AI has been around for a long time, but generative AI has kind of rekindled interest in AI. I want to talk about generative AI from two different lenses. One is observative for generative AI. That means people are doing all these generative AI related thing and how observative can help those systems. At the same time, generative AI for observative where you are leveraging a lot of generative AI technologies to further improve observability. How do you look at it? So one thing is there, so I really love technology and I'm sure you do too. We all are like really big technology bus, but there's also a risk here, right? That when we love technology so much, the question is, do we put technology before the use case and we all have to always keep that in mind that as much as we love technology, the bigger focus should be on the use case. So you're absolutely right. For Telma, also, there were two dimensions when we generative AI and AI has always existed, right? Like we all, just in my life span, I've done a lot of machine learning initiatives. November of this year with OpenAI and ChatGPT coming in, this just accelerated because it got more and more accessible to people, right? Like it's not just the big budget people who are doing, I'm talking to so many people, it's already on their roadmap. With Telma, we took a two-step approach. First is internally, where all can we use GenAI, whether it's in marketing, whether it's engineering, building test cases, also within our product feature functionality, which features can be used or developed using GenAI faster, cheaper, better, right? So use case first, but let's start adopting it. And we found many use cases, including as I mentioned, like building our test cases, how we do mean time resolution of problems, data observability problem. Can we aggregate data from different sources, from different logs and create email response, which is more actionable? So that's one dimension for sure, which Telma is definitely focusing heavily. The second dimension is we are at a very amazing spot where we are a prerequisite for any ML and data initiative, right? Even analytics area. If you don't have a good foundational data pipeline, your ML and analytics use cases are not going to give the right ROI. So at Telma, we did a research. We all knew, everybody knows there is a hypothesis that you need good data for getting a better GenAI result. So we did a research recently, where we ran some test and on if you preprocess the data a little bit and control the noise level of data that goes through your model training or fine tuning, there was 10 to 15% precision difference if you do that. So that means if you control the quality of the data, you are getting better precision on your GenAI initiative. Also, the cost is a big factor, because if you go and just push data and start training it without doing any prep work conditioning and so on, there's a tremendous cost of redoing that, retraining those models and so on and so forth. So we will continue as a company to invest in that and prove out that and educate the industry that GenAI is a shiny ball in the room, but let's have your data hygiene first. Security is one of the areas, although we don't look at that, term is fully doing securely whatever we do, but we are not a data security company, but security, governance, data quality hygiene, all of these are really the foundational pillars for the success of your GenAI and Telma sits very well in that story. I also want to talk a bit about the solution services that you folks offer. Give us a quick overview of what do you folks offer and if you can also just walk us through the evolution of your solutions over time, if there was an announcement and the updates recently, that would be great as well. Telma provides a full end-to-end solution around data reliability, if you may think about it, with the foundation of data observability and why I say it slightly differently, I can say it easily that it's a data observability platform, but what does data observability do to a typical customer? A, you can profile data, scan data, get insights about data across your entire data pipeline and get full fidelity of health metrics. The other thing we can do is we can help you with what we learn about your data, establish some data quality guardrails, rules, contracts, around data quality, which helps these two things, help with knowing what you already know, what we call as the known unknowns, the issues that can happen in the data. The third thing Telma helps is understanding anomalies in the data, which are usually very hard to find using validation rules or eyeballing sample data, which traditionally was done through stewardship. We use machine learning to detect anomalies, drifts in the data. So example, a certain company never had a revenue more than X million dollars and suddenly you see that the revenue has spiked by 40%. It could be a true positive, it could be a false positive, but somebody needs to look at it to see if that this was like just an additional zero and data engineering problem or it's a true, because these type of issues have impact on business right away. So we will help you look at these type of anomalies and resolve them. And last thing, surprisingly, every company still has a lot of data, which is in either legacy system or on systems, which are trying to migrate. Telma will also help you understand the metric monitoring system. So are your metrics drifting between the source system and consuming system? Are there any inconsistencies between, let's say your legacy on-prem databases and cloud data warehouse? Is that data is moving properly? Is there missing data? Has the data duplicates increased? Those type of issues as well. And the use case that I mentioned just earlier, like prepping your data for your machine learning profiling, we call it like data binning. So you separate data. Suspicious data can go for either like be parked somewhere else and go for either remediation floors or like reviews. But the good data flows in. So your pipeline is still kind of pretty clean. And then you have like this time to review. So there's a lot of stuff we do around data reliability to accelerate your journey on data reliability. And you asked about what we did new. So in all honesty, when we started Telma, we were like focusing a lot on like, how do we do like a product led growth bottoms up? And we quickly realized the tremendous momentum is where like the business impact is. So we moved to enterprises, like we moved more towards an enterprise, both from a sales motion, but also value proposition, which meant that we had to change our product direction. We had to support like bring your own cloud, right? If a client is already in a zero ecosystem, we wanted to make sure we deploy in their environment, in their account, and no data moves out of their account. So again, same thing for Google. GCP and Amazon. So we are now can be deployed in customers, all three accounts, all three key cloud providers. So that's number one. We added functionalities, which are more, I was talking to somebody earlier today, and they had hundreds of thousands of rules. How do they move to a fully autonomous, right? Like even when Tesla moved to autonomous driving, we still can't leave our steering wheel, right? So as much as we say that, let's do ML based anomaly detection, and they still want to keep their rules. So we created capabilities to support the rules that they have, so that they can build and manage their rules inside Telma. So capabilities like this, time travel capabilities, which means that, how much time should your machine learning model take to learn about your data? Instead of like waiting for that, can we look at historic data and predict anomalies and so on? So that accelerates time to value. So we did launch the last release, which was one of the largest release, or biggest release we have, which brought a lot of these capabilities, which specifically enterprises would embrace, and what we learned through our conversations in the recent past. Well, thank you so much for taking time out today and talking with the company, walk us through the whole evolution, not just of the company, but the whole observatories base. Thanks for all those insights, and I see you folks are doing a lot of great things. So I would love to have you folks back on the show and continue to talk about, the work that you folks are doing to solve a larger problem for a much larger ecosystem. Thank you. Thank you so much, Swapnil. It was a pleasure. And thank you so much to put the word out about the space, the problem it solves. We definitely need more of these conversations to spread about data observability and the problem it's solving for the general audience. Appreciate it.