Leverage AI in Impact Investment Due Diligence

Hey everyone, how's it going? My name is Ian Isaacs. I'm the head of global growth for the Founder Institute. Super excited to welcome you back, uh, on this Monday. Uh, it's August 7th. We have an amazing presentation as usual planned out for you. We have Dr. Vijayak and Brian here. I've known Dr. Vijayak for a very long time. Um, they're going to be talking to you about how you can leverage AI specifically for due diligence within the impact investment space. So Really excited to hear from them. I'll let them self-introduce themselves shortly. I just want to do some housekeeping things just before we get started. First off, as you can see, there's a chat button on the right-hand side. Please use that, communicate with everyone right now. Tell us where you're calling in from. I'm based here in Toronto, Canada. Would love to know where everyone else is coming in from. We've had about 230 people sign up from across the world. All right, we see Amanda from Seattle. Okay, we'll get people coming in. We have Vivian from Cologne, Germany. Let's go Nova Scotia. David from Nova Scotia. Okay, we get some people coming in. I'll do a quick check-in later. The other thing that I always tell people is that underneath the messaging button, there's a Q&A button. Please feed us your questions for the speakers in the Q&A button. Because at the end of their presentation, we're going to go into Q&A. I will use that as a means to bring up questions up, and you can also upvote other people's questions, and it becomes a little bit more interactive. All right. And then the last 10-15 minutes, what we're going to do is we're going to break for networking. So please don't go anywhere. You'll see all these virtual tables. Each speaker, Dr. Vijayak, is going to have it. Brian's going to have his own. Go speak to the speaker, ask them questions. You know, that's your time with them, right? I think that's it from a housekeeping perspective. We'll try to go till about the end of the hour and then wrap things up here. All right, I think with that being said, oh, the last but not least and most important thing is at the bottom of your screen you can see emojis. If the speaker is saying something intelligent and you want to give them some positive feedback, please use the emoji. To express that feedback, right? So that's just the little bit of fun we can have just to show a little bit of engagement. All right, we had— so just looking at the chat here, we have people from Cincinnati, Finland, Virginia, Bulgaria, Nigeria, Victoria, BC, Argentina, Phoenix, USA, Joburg, South Africa, New Orleans. All right, so we have a pretty international crowd here. With that being said, would love to welcome our panelists, Dr. Vijayak and Brian. Feel free to unmute yourself. Vijayak, I know that you have a presentation. Feel free to start sharing that and let's get things started.

Awesome. Thank you so much, Ayhan. I'm going to bring up the presentation so we can get started.

Vijayak, you want to tell everyone where you're calling in from?

Where I'm calling in from? I'm actually in Southeast Asia. I'm in Vietnam right now, but I'm usually based in Toronto, uh, when I'm not elsewhere on the planet.

Awesome. Well, thanks. And the only reason I did that, just to show that we have an international thing. And Brian, where are you calling in from as well?

Yeah, hi, uh, I'm in Denver, Colorado, USA.

Awesome. All right, cool. I think Vijay, your presentation is up and running. I'm going to remove myself, folks. You'll see me, I'll come back on to moderate questions. Uh, Vijay and Brian, show us, you have the show for you now.

Awesome, thank you so much again. Okay, great. So yeah, when, when I am first, uh, when, when I had the conversation with Ayan first a few months ago and the idea came up to come to this Founders Institute series. Uh, yeah, we were just at that point, Brian and myself, we had just, uh, really fully started to work on Syntegrol. And since then, a lot has happened. So actually, you know, we're— I'm thinking of this sort of, you know, we're in the midst of cooking up a venture that we're extremely excited about where we're solving some big problems and you know, and, and bringing something great into the world. And this is sort of like, you know, an opportunity for us to, to have, you know, to share a, a view into our, into our kitchen as we're cooking, basically. And, and so don't expect any fully developed, uh, and done recipes and solutions, but it's more like an in-the-making sort of process. So we'll be sharing about our learnings, our insights as we're building, uh, Syntegrol. And that's just for what to expect. And at the end, I think we'll open up a little bit some of the experiments that we're doing, uh, to potentially share with, with you guys as well. So I hope that this will be both a bit informational and educational, and maybe at the end a little bit experimental too as, as, as an event. But yeah, first I think like just to give you guys an idea of who we are That's us. So that's CitiRule. We're the co-founders, myself and Brian. Brian, do you want to introduce yourself first?

Yeah, sure. Thanks. And hello everyone. My name is Brian Streit and I previously contributed key technical IP to a current unicorn. I have multiple patents pending and I've previously built and led data teams at companies listed on the New York Stock Exchange. And I have a mathematical physics PhD. I had a bit of an academic background before switching over to industry about 10 years ago.

Great. Myself, as you can see here, I've had a varied background. I've led an incubator in Toronto for a couple of years where we did some damage to the Toronto ecosystem. A lot of great things came out of that. That's actually where I met Aihan, we actually hosted the Founder Institute event there in 2016. Was it Aihan? '15, '16, where we hosted a ceremony of award winners of the Founders Institute at that time. But I also led the digital transformation of the university in London before, and I've got a book coming up about futurism, technology, where the journey is going to. But at Syntegro, what we're doing is we're actually both here together focused on, on a very concrete task, and that's to create asset intelligence for a conscious economy. That's what, that's what our focus is, our mission, if you will. To start with an overview of our purpose, I think that's always good to know why people are together, what they're doing. We're basically building a global brain for sustainable and strategic asset intelligence. And our goal is to be the Bloomberg for the decentralized era. Now, what does that mean? Well, you know, strategic and sustainable asset intelligence. We're looking not just at financial data, we're looking at non-financial data. I'm going to get to that in a bit. Just how we're doing this, we're basically, we're building the world's first synergistically integrated intelligence platform combining real-time data, formalized knowledge, decentralized human insights, and AI. To deliver sustainable and strategic asset intelligence. And who are we delivering this to? Various stakeholders, players in verticals such as built environment, finance, and supply chain. Sounds like it's a broad spectrum, but actually there is people in these different areas that have very, very similar problems that we're solving. So this is kind of the big vision or the big purpose. It's always You know, I think any startup, if you want to grow and if you want to bring, you know, a lot of energy into it, you need what Peter Diamandis calls a massive transformative purpose, or, you know, the kind of Peter Thiel approach to having the big vision is really important. But then again, it's also really important to solve actual problems and not just be with a vision. So we're looking at solving very concrete problems as well, obviously, and those concrete problems They are in the area of delivering reports and intelligence. So intelligence in the form of reports on particular, on a very particular field. And that particular field is sustainability-related reporting, basically. So impact and sustainability, these terms are a little bit interchangeable. Actually, we, this is part of our journey. We first started emphasizing more the term impact. Since then, we have actually a little bit moved towards more using sustainability as a term. We can get to that why, but that's part of our kind of market discovery. That's been part of our market discovery as well. But putting that aside, the point or the idea is that you increasingly nowadays with the changing economy, with obviously with pressures from consumers and employees, and obviously the regulator first and foremost rising, there is an increased need in understanding non-financial aspects of assets as well. So it's not just anymore about understanding the financial performance, but people, investors, asset owners, asset managers, as well as the various other stakeholders, they want to know the more holistically, they wanna understand the performance of assets, So not just the financial performance, they wanna understand what impact is there on the environment, what impact is there on society. All of this falls under the bucket of sustainability or impact depending on where you are and which kind of terminology you prefer. And in order to ascertain that and find out, you need to read lots of different data sources. So these data sources could be anything from company documents, annual reports, sustainability reports, articles, news, both local and international, social media, laws, obviously policy regulations, real-time data providers, climate data, and human assessment. Very importantly, we also always need the human in the loop, the human assessment, or the human expert, subject matter expert, and their perspectives. So all of these different data sources, obviously to read them takes time. Analysts are going to spend a lot of time to read through these sources, and we need to, or we are able to cut that time down and create efficiencies there by doing that work for them through our platform. Now, an added difficulty or challenge is not just that there is so many different data sources to read through, but in addition, there is lots of different reporting standards. So we have impact frameworks or standards or sustainability impact standards, and each one of the different standards has different basically a different thesaurus or lexicon of terms and metrics and relationships between those metrics and rules. So it gets quite complicated. So if you're looking to understand the impact of a venture from a typical impact investor perspective, you have a variety of different frameworks, IRIS+, obviously the SDGs, which everybody knows, but there's lots of other ones, more obscure ones, competing ones. And in addition to this, 70% of asset managers who are interested in impact have their own proprietary frameworks as well on top of this. So you end up with really an almost infinite number of different frameworks that different people are using in different contexts. So all of these frameworks, that's an added complexity when you want to understand impact or sustainability. Obviously every industry has its own one, its own standards. In real estate, you have GRESB and so on, so forth. Every industry is going to have their own, uh, reporting standards as well. So that's the— that's the really the where the difficulty of this, of this whole, of this whole problem lies. Now, who needs these reports? Who needs our software to come and read the real-time data and then, and then understand all of the different reporting frameworks and then generate outputs in form of reports and insights that are nicely packaged that you can just simply download and integrate into your workflows.

Who needs this?

This is a good question. We actually initially thought that it would mostly impact investors, and this is where we kind of evolved a little bit because after, you know, very kind of intense rounds of market discovery or customer discovery, we found out that impact investors actually, even though they do have the interest, very high interest levels, they love to hear about AI solving their problems. They love to hear about AI, you know, working in the fields that they're working in, being applied to what they're doing, but they don't necessarily have serious pain points. They're like, they're just interested. So sometimes there is a disconnect between the interest levels. I'm just putting that out as well as a kind of a word of advice if you're doing initial market discovery or customer discovery and you're hearing, you're having a lot of great conversations with people, it doesn't necessarily mean that they have pain points. Maybe they're just really interested, but there is no dynamics or economics in the backend, in the background, which would actually give them the incentive to make it, or it could actually, they could actually be, their economics could actually be against you cutting down their time, right? So maybe because, that means that they have to cut down their teams if now an AI is doing the analysis for them. So we found out that there is some markets that are better or at least currently more ready and where the pain points are stronger, especially these are ESG consultancies and real estate firms which need to do sustainability assessments where it's really, you know, it makes a real difference for them to quickly read all the different data types that I outlined before to know the standards and sustainability standards that we also had on, you know, that I had up here, and then to generate reports. So we've had much more success and much more luck with— or it's not, you know, much more— much better results with these fields, with ESG consultancies and real estate firms who are utilizing our software in pilot engagements for generating reports on their assets. Okay, so that being said, now how do we actually do this? We've built, you know, we've built a what we believe to be a market-leading or a pioneering software in the field. We haven't found anything else that does this. And I can tell you a secret, it's not ChatGPT. It's not just ChatGPT. ChatGPT isn't enough, right? You can't just use ChatGPT to plug in your data and then to plug in the standards and then to say, okay, now give me a report, right? That's not going to generate the reports you want. What do we actually do? Brian, I think this is a good point for you to take up the baton and continue it from here.

All right, sounds good. Thank you. As Bichak was saying, we need a platform that provides real-time data with transparency into sources. That transparency is crucial. We want to have accurate knowledge of reporting standards and frameworks included in this platform. We needed to adaptively learn through human feedback. The reporting standards are going to change, reporting needs are going to change, the way that humans interact with this platform is going to change. We've seen that ChatGPT and Bard and large language models are great at generating text, but what about deep domain knowledge? Are they going to do your taxes for you? Are they going to extend beyond simple question and answer type interactions? And what about recent events? Uh, so at this point, it's prohibitively expensive to keep a large language model reasonably up to date on large amounts of internet data, right? So for example, you know, OpenAI's GPT-4 is, you know, 2 years stale now almost, uh, in terms of the data that it was trained on. And so how do we expand a large language model's knowledge base to include data that we care about. So when we get into real-time data with large language models, let's talk about what real-time data is. Okay, so data is called real-time when it is being used immediately after collection, and generally the cost increases as time between collection and use decreases. So for example, at a hospital, you're going to want real-time alerting in place for things like breathing and heart rate, right? But what about for due diligence? In order to get our news and social media and expert opinions loaded into the platform within minutes, right? How do we do that? Well, large language models and prompting. Okay, so One method for making real-time data available to a large language model is including this data in the prompt to the model as context. You can think of a large language model as something that has been trained on a lot of data, data on the internet usually, and as of a moment in time. The way that you get information out of this is by prompting it into giving you an output. So we can think of the prompt as an instruction to the large language model. And a contextually engineered prompt includes instructions, context, and a question. And the goal of the large language model here is to answer the question. And the knowledge of the large language model can be expanded into real-time or near real-time, right? We're not in the— situation necessarily where in a second we need to know if a heartbeat stops, but rather within a few minutes we need to have any relevant content that was pushed out onto the internet available to our platform, right? So including this real-time data as part of the prompt context. All right, so prompt tuning a large language model. When When asking an LLM to generate text, the size of the instructions is measured in tokens, but you can roughly think of these tokens as words. So the situation is, given a text generation task that you would like to be completed, for example, in the case of performing due diligence on some assets with respect to some framework, and a limit on the number of words that can be used to construct a prompt, How should a prompt be constructed in order to best complete your task? Now, if the text generation task requires knowledge of real-time data, then how can this data be made available to construct a prompt? Which brings us to current real-time large language models. So ChatGPT, it has plugins and there's some nice options for integrating real-time internet data with conversational AI. However, you can trip it up relatively easily. A simple question such as asking who is the CEO of some public company, you can run into problems. And this is because there's a lot involved in making real-time data available to an LLM. And so specific use cases are being addressed by those willing to build the infrastructure. And with these specific use cases comes subject matter expertise. Improving due diligence. So, so what does the due diligence process look like? Well, it at least involves some access to real-time data, right? Whoever is performing the due diligence can always just hop on the internet and you know, browse, pull data, whatever, right? And they have some knowledge of a framework with respect to conduct due diligence, whether it's, say, SDG or IRIS+ or, you know, any other sort of reporting framework. And so an LLM built to assist with due diligence should have at least access to real-time data and access to some knowledge framework. And even if the LLM were presented with all the documents defining a due diligence framework, it's still not going to do well with logical tasks. So this is where the subject matter experts step in. And you can in fact see how an LLM fails with logical tasks. It does recollection, recollection, right? So if you ask it some riddle and it's in its training set, right, it's going to be able to answer it. It's just recollecting, you know, some textual context behind your question, the riddle. But to actually perform logic, if you give it some setup, ChatGPT, for example, if you give it some setup and you try to step it through some really basic logical reasoning arguments, it's going to fail. So you shouldn't reasonably expect it to be able to reason with respect to some large framework of knowledge. So we have a need for formalized knowledge in this context. Since large language models don't do well with logical tasks, it's difficult to get large language models to analyze a set of documents with respect to information outlined in another set of documents. So an example from some experiments that my team did is simply loading SDG or IRIS+ documents as data, just the documents themselves, versus doing some intelligent prompting with them, uh, we of course saw much better results with the structured data. So essentially using these frameworks to intelligently engineer prompts instead of just making these frameworks available as documents for the large language model to sort of naively access in prompting. All right, so integrating real-time data and formal knowledge. So the goal is to analyze documents with respect to some formal knowledge and we have subject matter experts help define what this formalized knowledge looks like. Then the formalized knowledge is used to generate prompts that contain well-structured contextual information from the real-time data.

That's me.

Good.

I think maybe this is helpful for you guys to just get a look into how some of the technical things that we've been grappling with in our first number of engagements and pilots that we've run. We've basically, as Brian was explaining here, we've basically brought together the capability to read the real-time data, whether it's on the news or social media, Twitter feeds or X feeds, whatever, or or, you know, relevant data such as, such as climate or other relevant data for that particular asset class, plus, plus any type of proprietary data around the asset, plus the formalized knowledge of particular standards, reporting standards. So in the case of impact, for it as an example, in the case of impact investors, Impact investors are interested in impact themes. So here are some sample themes. For instance, they're interested in access to energy or mental health or women's empowerment. So these are some sample themes that they're interested in. And each one of the themes is going to have certain metrics and certain rules and certain objectives. And it's going to be connected in some ways to the other teams. So that's an example for the type of formalized knowledge that we have built and that we're able to add to the ability to read real-time data.

Right.

And when you do that, then you can have basically— then you have a due diligence solution, then you're able to generate outputs. Now, as I also said, throughout our journey, we started with impact investment. And that's where we wanted to first be. But because of the, the dynamics of the markets that we've explored, we've actually built out the solutions that we've built out over the last couple of months are actually in other fields. They're mostly in ESG consulting and real estate, uh, largely. So those are other fields that are not actually impact investing, which is why, you know, our talk, which was set— the title was set a couple of months ago, and we've actually not that much, you know, worked on impact investment. We're more working on The kind of— it's related sustainability certification and sustainability assessments, but it's not exactly the same, right? However, we are very interested in this field still, and we want— it is related, obviously. It's close to each other. They're close adjacent fields, even though slightly different. Maybe in the Q&A we can go into this. So however, we have a, you know, a potential use case which we'd love to explore maybe with some of you. Our technology is able to read real-time data. We can read the news, we can read feeds, social media feeds, we can read any data rooms that any asset has, such as your startup. For instance, your startup is an asset as well, is an asset for any investor who wants to invest in it, right? An investor looks at your startup as an asset, you know, with a particular kind of value and a particular ability to due diligence it and some financial as well as some non-financial values, right? So that's a use case that we're very interested in. So if you have a startup that you're, that, you know, that's yourself, or if you're an investor and you want to look into a startup, what we can do is, as a kind of experiment, we can load up your data into our, into our system, and we can generate some outputs on the impact that your startup is, is either generating or that it's or that it's prone to generate. Obviously, if you haven't launched yet, or if you're, you know, if you're an early-stage startup, there's not gonna be that much data, which is actually one of the big issues that impact investors have that are looking for startups. It's so difficult to predict if this startup is actually going to have that impact. But what you can do is, and I know that impact investors do, you can look through research, you can look through scientific journals, you can look through evidence portals, Those sources will tell you, for instance, that, okay, these guys are building a certain application and this application helps with productivity or helps with mental health. What is that going to look like? What will be the impact on this? Are there any studies that are backing this up? If so, then can we go there and can we find them and can we ascertain what impact categories and themes and with what confidence levels this startup or their venture or the product is going to contribute to this particular impact theme, right? So that's an example, but there's many others. So if you're interested in that, go on the link bit.ly/syntegral, and it's a page there where you can just express your interest, uh, as, as a startup who wants to take part in this experiment with us, or also as any other collaborator or other party that may be interested in due diligence. Again, our actual focus is currently more on, on real assets and on some complex kind of more complex assets that require sustainability assessments where we have to look at ESG factors and so on and so forth. But our technology is perfectly capable of handling this as well. And we'd love to have that experiment, maybe learn from it together with you guys. You can learn, we can learn. We can all take away something from it. So if you're interested, go on the link and, and, uh, you know, and put down your digits and we'll talk. And I think now we're open for Q&A and any questions and any comments and any kind of open conversations.

Yeah, thanks everyone. Uh, thanks Dr. Vijay, thanks Brian. Let me just— I'm going to stop sharing your screen here. Everyone, let's just head over to the Q&A. I see Shah just killing it, added like 10 questions in here. Let's do a few of these. I'm going to bring up the first one by David. You guys should be able to see it. David from Berlin's asking, do you use generative AI for deterministic tasks? I don't know, Vijay or Brian, do you want— whoever wants to take this?

Yeah, I can take it. So, um, for due diligence of an asset, um, uh, you'd wish that, uh, somebody doing the analysis of, you know, an asset trying to answer a specific question, uh, they would get the same answer every time, right? And in some cases, for example, like soil contamination on a commercial real estate property, uh, the evidence is pretty straightforward and they can come to the conclusion relatively easily. But some questions are a bit more nebulous, right? But our AI system doesn't even try to automatically answer the questions for you to make some sort of business decision. It gives you its answer and allows you to do the QA yourself, right? That's part of the transparency aspect of the platform. So, you know, it gives you some answer to a due diligence question, right? And you can go research that answer yourself very easily because you're hyperlinked directly to the document subsection that contains the information relevant to that question.

Okay, perfect. Let's kind of move on here. Greg, can you see—

I'll just jump in there. Quickly?

Yeah, for sure.

Sorry, I had shared a link and it looks like the— I, you know, I use the Bitly like a link shortener and it looks like that didn't work. So somebody was right there, it didn't work. So I just shared the unshortened link as well. So yeah, the link that I shared on the thing was not— that was—

yeah, I just threw it up for everybody to see on the chat. So it's on Yeah, thank you. No worries. Let's move on. Greg had this really good question. Given that social media, press releases, and other biased sources are part of the training set, how do we defend against greenwashing tainting the results? Are we able to identify the degree to which a given org or fund is greenwashing? I guess you can take it a step further. Like, there are a bunch of different data sources. You don't know whether they're accurate. You don't know if they're correct. So how does a product like yours, or just in generally in the industry, like, can you take a data source, like, should you clean it first, should you verify it, and then put it into a product like yours to pump out the respective results? Or are you guys doing that automatically? Would love to learn more.

Yeah, this is a great question. This is a great question, uh, Greg. So that's exactly— so if you're, if you're just going to any vanilla LLM, and obviously they're not trained on all the most recent and relevant data, right? They have a certain cutoff date and they're also trained only to the limited sets of data. But what's really important is what you just said, you know, there is no distinction between the sources of some information in that training set, right? So, and that's something that we have created in our platform. So In our platform, the platform is going to distinguish between the news and the social media mentions and the self-reported company documents, right? So they're gonna be different. So they're not gonna be treated as the same kind of information. It's gonna be distinguished and in your outputs it's gonna say the company reports and says such and such, but the news or, you know, third parties are saying such and such, right? So that's a really important point because obviously otherwise, if we're not making the distinctions, there is the, the problem of greenwashing, impact washing. I'm not saying that this is already solved just by making that distinction because it's more complex than that, right? So it's still, it's still, uh, there's still work to be done to really, you know, solve the problem of greenwashing or impact washing. But one step towards that is to have you know, clear provenance of where data is coming from, that, that, you know, where, where the, there, where the data is coming from for a particular output that you're, that you're generating.

Perfect. Brian, anything to add, or can I move on to the next question?

Yeah, I'll just echo that this all comes back to the data transparency, right? What, what is the data being used in the process of constructing an answer? Answer to your question, right, within the context of large language model.

Okay, perfect. We got a bunch of questions here from Chad. I'm going to pick up a couple just to make sure. I think, um, here, so Chad's asking, you know, does the AI give you better perspective on global demographics, economic trajectories of certain areas.

Global demographics and economic trajectories of certain areas. Can you elaborate on that?

Hey, Shad, I'm gonna bring you up on stage. Let's see if— oh, okay, you've joined from your mobile. You can't think. Um, okay, Shad, can you— if you can add a new question of this. I think what the person is asking here is like, when you put in all this information, can the AI give you a different perspective on like upcoming global, upcoming global trends and help you identify different demographics and economic trajectories, right? So like, especially specifically with, I guess, within the impact space. I could be wrong.

I mean, look, there's— we've really focused on solving the due diligence problem, right? So the specific problem of there is an analyst, due diligence professional, they need to read this data, they need to put some of their own thinking into it, they need to know the reporting standards and they need some outputs on that, right? So that happens whether you need green building certificates on buildings right, which needs certificates that need to be renewed from time to time, and that's mission critical for, for the real estate sector, or whether you need certificates in supply chain, or you need certificates on, on, uh, you know, you know, on es, es, uh, factors on environmental and social, uh, sustainability due diligence. There's some certain, you know, several different, uh, frameworks there that exist for companies. So It's a similar process in each case, okay? And that's the problem we're solving. Now, could AI be solved to use to, you know, obviously, yeah, there's all kinds, for instance, you know, like to bring this down to a building again, because we're working with real estate, right? Our solution would not be able to predict the climate. There's others that are building digital twins of the entire kind of climate system or of the entire atmosphere. That allows you to predict, you know, what the weather will do, what the kind of climate changes will do, and then draw inferences from that. Our software doesn't do that. We just do due diligence on existing datasets, right? So just like this, we can do due diligence on reports that exist on demographics, but we're not able to, we haven't built a digital twin of the global population or any kind of predictive engine that tells you about how the population is gonna grow.

Got it. Okay. I think that's helpful for people to know, not just for your product, but for just in general how these products work is like you're feeding, you're feeding the, your product or any other products with a certain amount of inputs and the output is going to be very, very maybe singular or might be something parallel, but it's not going to give you a full landscape unless you really work it in. Just to move on, I'll do one more question and then we'll break out for networking. This question from Igor from Panama is pretty interesting around data security. So he's asking, since you're dynamically loading LLM with real-time data and prompts, do you have data security concerns? Meaning once the data is in, it's available for others to get access to it with simple prompts. I think if you can extend this beyond your product, but I think this is a general concern with a lot of other AI products out there. I know my brother works at a European company and they're not using ChatGPT at all because of this, for the very reason they're going to wait for Microsoft's AI platform because it will have better data privacy policies. So would love to get your thoughts on this as well.

Brian, is that one for you?

Yeah, I'll address this. Uh, yeah, I mean, we address data privacy concerns as they come up, right? So for example, doing business in the EU versus doing business with, say, like, you know, HIPAA data or governmental data, these all have their own restrictions. And if it requires, you know, on-prem with a locally hosted large language model versus cloud infrastructure, that's something that we're in a position to address. But yeah, it's definitely a concern. It's a good question. And it's definitely not something to be skirted over.

Yeah.

So, and I saw that as well. There was several people before who had asked similar questions and absolutely very, very important question. And as Brian was just saying, we have solutions for every level of requirement of privacy and security that we can implement on a needs basis.

Okay, perfect. Yeah, thanks. I think the data privacy and AI portion is very, very interesting. Okay, we have about 108 people still online. What I'd love to do is make sure, you know, we have 15 minutes left. We're going to stop the presentation here. Brian, Vijayak, thank you so much. This has been very interesting, very helpful, very different from the usual conversations we have. We'll break out for the— everyone will see the virtual tables. Vijayak, head over to your table. Brian, head over to yours, and then you can meet some of the participants. But thanks everyone for joining. Vijayak, Brian, any last words for anybody that's just about to hop off?

Yeah, like, you know, it's a bit, you know, every— I see that there's a lot of people there, but since it's a bit different to be in an actual conference or being actually see people mingle and so on. So yeah, I hope to see some of you people afterward in the rooms or elsewhere on LinkedIn or elsewhere. And you know, there was lots of comments here which I think would warrant interesting conversations, so hopefully we can have them elsewhere. Thanks. Perfect.

Brian, any final words?

I'll just say thank you for the opportunity. It's been great.

No, thanks for coming. We love hearing about companies like yours. It's very, very interesting. I like how you're combining AI, sustainability, and solving these new problems. Um, just, uh, because it was asked, uh, yes, you will receive a recording within 24 or 48 hours. You will get access to everyone's LinkedIn and you'll see the deck there as well. I'm going to end the session here. Thanks everyone. Enjoy the rest of your day and then we'll see you in the networking. Bye.