How Permutable AI Is Turning Unstructured Data Into Trading Insight

[00:00:04] How do you make sense of financial markets when headlines move prices faster than balance sheets ever could? Well, my conversation today is about that tension between data and narrative, and how AI is reshaping the way that financial institutions might interpret both. And my guest today is founder and CEO of Permutable AI.

[00:00:27] And he's someone that began his career as a trader before leaning fully into computer science and machine learning, long before it became fashionable. And that mix of market instinct and technical depth, these are the things that sit right at the heart of what he's building today. Because permutable AI is something that focuses on real-time intelligence across commodities, energy and currencies, scanning hundreds of thousands of global articles and information flows,

[00:00:57] putting them all into one big melting pot to understand how the narratives are forming, spreading and ultimately influencing price action. So today we'll explore the shift from traditional quant finance to what Wilson calls AI native software 3.0. And by that, he means systems that reason over unstructured global data, rather than relying purely on structured financial inputs.

[00:01:25] And he'll explain why vertical large models built specifically for markets can embed financial relationships to ontology directly into their reasoning, producing outputs that are both structured, explainable and traceable, all the way back to the source data. And yes, it is the year of agentic AI again, isn't it? So we'll also talk about multi-agent architectures, adaptive world models,

[00:01:51] and this idea that in modern algorithmic markets, perception and reality are effectively beginning to merge. So expect a fascinating look today at how AI is evolving from just a tool that analyzes history to a system that can model the present in real time. So if you care or just interested in where capital markets, AI and narrative intelligence intersect, you're going to love this one.

[00:02:18] Let me officially introduce you to Wilson right now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? My name is Wilson. I'm the CEO and founder of Permutable AI. We are a, obviously the secret's in the name, but we are an AI company.

[00:02:43] We focus on market intelligence around the world for financial institutions. And my background is I'm an ex-trader, but I had a big passion for computer science many years ago. So this is like a perfect project or company initiative for me to do. Yeah, it sounds like the stars have aligned for you there. And I love your backstory and how it's led to where you are now.

[00:03:09] Permutable AI was founded on this belief that markets have moved from those traditional quant finance to AI native software 3.0 systems. I've got to ask that. What was it that fundamentally changed in financial markets that made this shift inevitable? Because you were right at the heart in this space. You saw what was happening and made your move. But what was it that changed?

[00:03:33] I think I was doing a lot of early machine learning type work for the US investment banks 15, 20 years ago. And it was really clear to me that eventually this type of work is going to get automated, is going to evolve further, and it's going to propagate its way through to other sectors.

[00:03:53] So, yeah, I mean, even like 15 years ago, we, I saw how powerful that some of these early machine learning models would be, and how they can sort of generalize pretty well to different tasks. And I think what we're seeing now is that that rate of acceleration is the highest it's ever been, like even over the last 12 months.

[00:04:17] So I think what we're seeing now is what people used to call the sort of singularity event for AI, where it's improving at such a rate that it's almost kind of unworldly. So I think that's what's happening now in the middle of it. So, yeah, we're in the middle of a very exciting sort of transformation in this sector. And before you join me today, I was doing a little research on you,

[00:04:42] and I was reading that you said that the large models unlock a new class of market intelligence and do that by reasoning over unstructured global data. So how does interpreting narratives, policy shifts, and real-world information flows, how does this change the way that investors make decisions, just for anybody in that field that are listening today? So a really interesting discovery that we found with LLMs is a lot of people traditionally used to call LLMs

[00:05:12] just a simple form of regression on like historical data. So what those people are saying is that if the LLM sees anything new, it doesn't know how to handle it. It wouldn't know how to respond. It only can respond if something has happened in the past. What we're seeing with the new LLMs is, especially with the reasoning layers, is that they are able to deal with new scenarios. Okay.

[00:05:39] And that's like a critical difference between like what is a classic like regression model versus something that actually has some sort of contextual understanding. So we're seeing that now in latest models. And I think that's going to really, really help the market, especially in these times of world order where we're seeing events for the first time. Thanks to Donald Trump. There's a whole other podcast right there. But obviously, I mean, what makes you stand out?

[00:06:06] You build vertical LLMs purpose-built for markets. So what is the advantage of embedding financial antology and relationships directly into the model rather than relying on just general purpose AI? So models are quite good. They can give you really good answers. But what we care about is like full traceability to data and all the different layers of thinking.

[00:06:29] So in an output, you can see layers and layers of talking and thinking and rethinking. That's quite useful. But we'd like to actually trace back everything back to like raw data for every layer, if possible. So what are the steps? Like what are the hard steps that you made? So with the kind of knowledge graphs that we build,

[00:06:55] it really allows us to provide that kind of auditability, that kind of traceability for our clients. So yeah, if you let LLM think too much, it does a ton of ton of thinking, you get an output, but it doesn't really help you fully understand how it got there. And I think that how it got there is almost just as important as the final output as well. And for any investors listening to our conversation today, one of the exciting parts of what you do here is

[00:07:25] your platform scans hundreds of thousands of global articles, all in real time, and measures how narratives are beginning to form and ultimately influence prices. But I've got to ask, how do you separate signal from noise in such a high volume information environment? I would imagine that's one of the biggest challenges. So how do you do that? Well, you're kind of asking us, what is that? You're kind of asking us, what is part of the secret sauce as well, right?

[00:07:50] So we built models and architecture that allows us to really deal with the high amount of volume of noise coming through. And obviously, if you start sharing that noise to clients, then it's not really a great product. So we spend a lot of time actually working on the noise filtering at the lowest end of the stack. And obviously, that's the most intensive as well. And that can be, if anyone else was to try and do that, it would be very, very expensive.

[00:08:19] Yeah, I love it. And you also describe your system as building a continuously adaptive world that maps assets, economies, policies, and narratives as they interact. So what does that look like technically? How close are we to a true capital market simulator of sorts? Are we close to that, do you think? We've studied all the, what we call point-in-time shifts. So what point-in-time means is,

[00:08:47] are you replicating the past exactly as it was, like based on the data and based on the models that existed at that time? How well of a time machine can you create, like based on your framework? So because we've spent a lot of time doing that over the years, it's something that we can evaluate how good the models are at any point in time. So for example, I'm obviously talking about Trump over the last 12 months,

[00:09:15] but what if there was another new event that happened in the past? How would you evaluate how good the systems were at dealing with that? And if you can go back in time and actually show that there was a model that existed at that time that you could have created, how well did that do at that time? So I think it's evaluating those points. And if we take it a high level sort of further, we want to know if the next LLM can create the cure for cancer

[00:09:44] or if it can create like the next technological leap, right? That's what people are hoping for. You can kind of test that out already because you can take a model, you can bring this model into the past and see if it can actually reinvent the internet, for example. So this type of work is actually pretty critical and it's going to be more critical in the future.

[00:10:11] But this is the kind of work that Permutable focuses on. And it does feel this year that there's a big narrative around all things agentic and agent related and multi-agent architectures are central to your approach too. So how do these collaborating agents replace or maybe even outperform traditional analytics stacks that are used by systemic funds that you may have encountered throughout your previous life as well?

[00:10:40] Yeah. So if you look at what's happening in the software market, the bottom two layers of software engineers are almost out of work, right? Or let's say 90% out of work. So that's going to move its way upwards through the chain. So what that is really telling you is there was a massive efficiency gain, like in the labor market, in the digital space. We're going to see that in other verticals as well.

[00:11:07] But what's quite clear to me is that we're not going to see a full displacement. So what might happen is you've got like a team of five people on a desk that is like managing money. Money, that will probably come down to two, let's say, for example. Okay. And those two would be very AI savvy and would be using agents to do work on their behalf. Okay. So, and then also the number of desks in total would probably shrink as well.

[00:11:36] Because you can imagine that you've gone from a team of five to a team of two, but the team of two is now doing the role of a team of 20, let's say, right? So you're going to get displacement for sure. But what's clear is that it's going to be a combination. Well, in my eyes, I think it's going to be a combination of agents and highly sophisticated humans. Like you still need that accountability. Yeah.

[00:12:02] And I would imagine that trust will also remain a major barrier to AI adoption in finance. So how do things like explainability, traceability, and auditability, how do they become competitive advantages rather than compliance burdens? It's probably a question you get a lot. Well, you have to assume that at some point the agent's going to do something wildly wrong. Yeah. And when something goes wrong, then what's the next thing you're going to do?

[00:12:32] You want to know why, right? Yeah. And if you don't know why, then you lose complete trust of the entire system. So if you can't trace back on why someone has done certain steps and you can't trace back the source of error, then the trust in the whole system collapses. So that's why you need it. Like I'm pretty sure if you don't have it, then all trust will collapse. And I'm curious, what kind of feedback have you had from the finance community?

[00:13:01] Obviously, you know a lot of people in this area anyway through your career, but when you're talking to clients, do you get the same kind of questions? What kind of feedback have you had? It's a really interesting dynamic because it's clear that the technology is really, really advanced and it's getting more advanced now. But the financial sector is a sector based on legacy, like based on trust, right? It's based on reliability.

[00:13:29] If that gets compromised, then you're compromised in the sector. And I think the trust issue of agentic systems is still very questionable. Like no one who, like would you trust an agent with dealing with your medical advice, right? Like would you, would you trust one to prescribe you pills, you know? So if the answer is no, then that's, that, that's the issue, right?

[00:13:57] And then what would it take for, for that hurdle to cross? It takes quite a big shift. So I guess my point is when it really, really matters, then the adoption rate is good. It's going to come with a lot more resistance. Yeah. And if perception and reality have effectively collapsed into one, into one in algorithmic markets,

[00:14:23] what does that mean for portfolio managers that are listening who are still relying primarily on structured financial data and backward looking indicators, et cetera? So what do you say here? So, so my, my point with that is I think a lot of investment thesis is based on what is like the fundamental value of an asset. Okay. And the fundamental of an asset usually is based on how much cash it generates.

[00:14:52] What is the trajectory of that cash generation? But let's say, but we're seeing a lot of examples where let's take open AI, for example, it's losing. The reality is it's losing a ton of money, but people believe in the narrative. It's the same with like Elon Musk companies, right? They have really high multipliers because people believe in the story. They believe in the narrative.

[00:15:17] And our view is that if people believe that narrative is true, you might as well assume that the narrative is true. And you trade it because that narrative can persist for a really long time. As someone that's right in the heart of this space, a question I'd love to ask you is what do people most misunderstand most about your industry? Or are there any myths or misconceptions about your job or field of expertise that we can finally lay to rest today? I suspect you're someone that spends a lot of time online.

[00:15:47] You're in the various forums, Reddit communities and LinkedIn. Do you see any myths or misconceptions that frustrate you that we can just lay to rest today? I think what's really, really surprising is that the rate of adoption and rate of innovation in the financial industry is nowhere near where it should be. And that's really, really surprising because they have the smarts, they have the capital behind them

[00:16:16] and they have the need, right? But it's nowhere near close to adoption rates as you'd think. You'd be very, very surprised how many people still use an Excel sheet in the finance industry. So I think that the bottleneck could be the people, I suspect. People just not being skilled to really embrace this.

[00:16:43] I think that's probably the bottleneck and that probably explains why the tech industry has innovated so quickly because they're effectively a group of programmers. And as for yourself and everything that you're working towards this year, you've probably gone to a lot of conferences and things as well and a lot of deliverables throughout the year, which you probably can't share too much about. But what excites you about the rest of the year? What are you going to be working on? And what are you hopeful for this year?

[00:17:08] I think we've gone through like a 4x revenue increase like last year. So we expect something similar, like maybe 5 to 10x increase in revenue. So what that tells me is that people are very open to, as long as they get shown how to do it, we think that this transformation curve is very real. And we at Permutable AI, we're slightly different to like a consultancy in that we're actually practitioners

[00:17:35] showing people how we've done it and how we build these systems and put them on-prem. So we're really just sort of helping, you know, as much as we can, the industry to do these sort of transforms. And as this is a tech podcast, a question I've got to ask before you go, because I know it's a big part of what you're doing here. And it's probably a big question of what you get asked a lot is, when getting that adoption, first thing people are going to be thinking about is integration problems.

[00:18:03] And I know a big part of what you do is, hey, we offer zero integration headaches here. So tell me a little bit more about that for anybody that might be listening that might have those concerns. The integration is actually the key barrier for any company looking to do like transformation projects with the FIs. We focus, we're quite similar to Anthropic in that we focus on the API part of plumbing. So we focus purely on data lakes, like data integrations,

[00:18:32] and we feed a lot of our AI insights through there. And then we apply that through all the various capital market sectors and assets. So that's a lot more friendly. So machine-to-machine integration is definitely more friendly. The bits that come later will be more the human-to-AI integrations, that they require more thought about how to do that. And I think that is a great moment to end on.

[00:19:00] But before I let you go, there's going to be a lot of people listening, curious all around the world, how they can find out more information, dig a little bit deeper, et cetera, find out more about new announcements. Anybody interested in anything we talked about today, where do you want me to point them? Yeah, sure. So just look us up, permutable.ai. You'll find us in the news, and you'll see all the announcements that we've made over the last two months.

[00:19:25] And that should give you, good people, a very strong idea of the kind of projects we've been working on this quarter. Yeah, and I think anybody listening that is interested in specialist LLM tools that deliver instant market intelligence, with most importantly, as we just said, zero integration headaches, I'd urge them to check you out. I will add links to everything you mentioned, along with a few of the big announcements, and your LinkedIn profile as well, maybe, so people can reach out and ask you any questions.

[00:19:53] But just thank you for bringing this to life today. It's such a big topic right now. It's evolving in real time, and I would urge anyone to check you out. But thanks for joining me. Yeah, no worries. All right, thanks. Thanks, Neil. I think one of the things that struck me most in the conversation with Wilson today is this emphasis on explainability. Because in financial markets, trust is everything. If an AI agent makes a decision that moves capital, you need to understand how it arrived there.

[00:20:21] So traceability and auditability, they are the foundation for adoption. So it was good that we touched on the hard truth today that the rate of AI innovation is accelerating rapidly. But many parts of the financial sector are still leaning heavily on legacy tools and even spreadsheets. But this gap between what is technically possible and what is operationally deployed may become one of the defining competitive divides over the next few years.

[00:20:51] So certainly something that I think we should all be watching closely. And if you are curious about adaptive world models, vertical LLMs for markets, or how multi-agent systems could replace traditional analytics stacks, remember, head over to permutable.ai, explore their latest announcements. And as always, I'd love to hear your perspective too. Are we slowly moving towards a true capital market simulator?

[00:21:17] Or are we still in the very early chapters of a much bigger story? Please head over to techtalksnetwork.com. You'll find 4,000 interviews and all the links to everything that I mentioned today too. So feel free to send me a message and I will also return again tomorrow with another guest. So hopefully I will speak with you all then. Bye for now.