Superposition Guy’s Podcast — Yianni Gamvros, CEO, and Iordanis Kerenidis, CTO, Quantum Signals
“The Superposition Guy’s Podcast”, hosted by Yuval Boger, Chief Commercial Officer at QuEra Computing
Yianni Gamvros and Iordanis Kerenidis, co-founders of Quantum Signals, a new startup focusing on B2B software for financial services, are interviewed by Yuval Boger. Yianni and Iordanis discuss their strategy of building classical AI pipelines to deliver immediate value to customers while planning to integrate quantum enhancements as the technology matures. They explore the challenges of optimizing large financial transactions using predictive signals powered by transformers and quantum-inspired methods, contrasting their approach with traditional trading algorithms. They discuss their initial successes, a quantum hackathon they are organizing, and much more.
Listen on Spotify — here
Full Transcript
Yuval Boger: Hello Yianni, hello Iordanis, thank you so much for joining me today.
Yianni Gamvros: Thank you, Yuval, it’s great to be here.
Yuval: So who are you and what do you do? Yianni, maybe you go first.
Yianni: So yeah, so we recently had a few changes and updates in what we do. So my name is Yianni Gamvros and currently I’m the co-founder and CEO of a new company called Quantum Signals.
Yuval: And Iordanis?
Iordanis Kerenidis: Yeah, I am Iordanis Kerenidis, I’m the CTO of this new company called Quantum Signals.
Yuval: And what does Quantum Signals do?
Yianni: So Quantum Signals is a new startup, so we had a new vision of what success might look like. And we decided to put together a new company to go after essentially this vision. And so we have essentially three founding principles. One is to concentrate on building B2B software. The second one is to go after a specific use case in a specific industry and that is financial services and that’s going after a specific use case in financial services called large order execution, and we can talk more about that. And the third is to start offering essentially a product and a service using classical AI methods and create basically a classical AI pipeline that delivers value to the customers right away. And then slowly add quantum into this pipeline as quantum matures and becomes more powerful.
Yuval: So you should have called it quantum-inspired signals essentially to start with and maybe you drop the inspired at some point later. Is that about right?
Yianni: It is about right. So it is about right that we want to reverse the way that typically I think software companies and application developers in the quantum space try to do things. Which is essentially take a very small problem and start with a very small toy problem and slowly start increasing essentially the size of the problem.o try to fit that into a larger and larger quantum computer or try to create a quantum method that can deal with a larger and larger problem. So we’re starting kind of with a reverse, we’re starting with a real life problem and then we’ll try to slowly inject essentially quantum into that computational pipeline.
Yuval: As someone who deals with sales, large order execution sounds really exciting, but maybe you can explain what it really means.
Yianni: Sure. So maybe I’ll cover sort of the business case and then Iordanis can talk about essentially maybe a little bit more about what exactly we have in mind and what goes under the hood. So large order execution is a problem that appears in finance quite a bit. n some sense it’s one of these holy grail problems. In finance where you have these large institutional players who are trying to execute these large orders that are maybe tens of millions of dollars. hey cannot actually put these orders onto the market as one big monolithic order. Other people are going to take advantage of that and try to move the price different ways. So they actually try to split up the order into many, many smaller orders and then execute essentially that sequence of smaller orders within essentially the trading day. And so the whole concept of large order execution and doing it more efficiently is essentially scheduling the smaller orders and optimizing the timing for the smaller orders, so that you take advantage of the change in price throughout the day and you execute with essentially the most favorable price. You might essentially accelerate the next small order or you might delay the next small order to take advantage of some short term price movement. So in a nutshell, this is kind of the high level value proposition. And then I don’t know, Iordanis, you want to start talking a little bit about kind of what happens under the hood and what we have in mind of how to actually do that in a smart way.
Iordanis: One way of trying to look at this problem is basically it’s a game between different people. So someone wants to sell a number of stocks and then some other people want to buy the stocks, but one’s trying to sell at the best price and the other is trying to buy at the best price. So the whole point is if you know that I want to sell a lot, then you will pressure me to sell at a lower price and so forth. So basically it’s a game between traders, between agents. And this is kind of the scenario where AI is very good at dealing with. So we have seen what AI can do with playing more and more difficult games and finding the strategies for these autonomous agents to do it themselves. And this is what we will try to do in the long term to have these trained AI agents that can take the best decisions in order to maximize the profits. And in order to train these agents, the first thing we can do is basically try to give them some advice. So the advice will be a signal that will say that, okay, in the next five minutes, we are predicting that the price is going to go up or that the price is going to go down. And then the agent can take this into account while we’re optimizing the strategy of when they are going to put the orders and so forth. So the signals is kind of what we are doing right now and basically in order to be able to predict what will happen in the future, whether, for example, the price or some other quantity that you care about, like the liquidity is going to increase or decrease, we are using basically these very new techniques, which are also the basis of the LLMs, which are called transformers, and we’re using these particular transformers that are for time series data because the data in finance changes over time. So you need to keep track of how things change over time. And we’re using these temporal fusion transformers, we are injecting some quantum-inspired ideas, as you said, already on them and try to get as accurate signals as we can to give to the traders, basically.
Yuval: So let’s run through an example because I’m curious to understand how is it done today. I’m a broker for Warren Buffett and Warren Buffett decided to sell $100 million worth of Apple. So it’s a large order that needs to be executed. I want the best price. Before you arrived at the scene, how is this order executed?
Iordanis: There are different ways of trying to do it. But when we really talk to many of our clients like banks or small hedge funds and things like that, what we find out is that the most usual way of doing it, it’s also a very simple way of doing it, which is to say that if I have a large amount of stocks to sell in the next five hours, then I split my order into 10 equal parts and I do it every half hour. So this is one of the canonical methods called TWAP. The other is not according to time, but according to the volume of how much has been traded per day. You try to follow the same volume and again you split it and you try to spread it along the day in a way that you are not going to impact the price negatively. So these are the two canonical ways. They’re also very optional ways. So this is what most brokers will use or refinements on these two methods. And one characteristic of these two methods is that they are not using any predictive signals. So you are not trying to predict what will happen in the next half an hour in order to change your decision. So this is where the machine learning and the AI comes into the picture where we add the signal, these predictions of, you know, maybe in half an hour the price is going to be better, so you should wait a bit and do a bigger order later on. Or maybe you should run because the price is going to go down. So adding these predictive signals into these more canonical and basic, I would say, ways of doing large order execution, this is what we think will give a huge edge on how to learn.
Yuval: And this is true both for buying and selling, right? It sounds like a symmetrical problem.
Iordanis: Yes.
Yuval: If you are able to give signals on the trend, if the price is going to go higher or lower, why is that relevant only to large orders? I mean, I could think about any order in any commodity that I would want to know if you think the price is going to go up or down.
Yianni: Yeah. Maybe one thing to think about here is the fact that we’re only claiming to be able to do, at least for now, short-term predictions. And so these are predictions that in practice will be anywhere from, let’s say, a minute, maybe all the way up to 5, 10, 15, maybe 30 minutes, right? So yes, you could say that some very busy intraday traders could take advantage of the signal and turn it into some kind of strategy. And in fact, actually one of the prospects we’re currently talking to is thinking about doing that. So yes, so large order execution is essentially one, is the central use case, but the signal we’re producing to power this execution can also be used to train systematic strategies. So yes, there is already some truth in what you said, and we’ve already seen that from the feedback we’re getting from the market.
Yuval: Would the signal tell me if I should wait or sell now, or would it also impact the size of the block that I’m trying to sell or buy?
Yianni: Iordanis, you want to take that?
Iordanis: Yeah, the signals can be as complicated as you want, as long as you train them to be accurate enough. So let’s say it is easier to have a signal that will just tell you, is it going to go up or down. It is more difficult to train a signal that will tell you what action you should take. So this is a whole strategy, and the signals will be put to this strategy. And then the strategy will train to take actions at every minute. For example, I look at the signals, I look at what is the best strategy, and then I take an action that says you should buy this amount of the stock or you should sell this amount of the stock.
Yuval: What is the status of the algorithm or the company now? Does it work? Have you tried it? Are you engaged with customers?
Yianni: Yeah, so actually good news is that we’ve already signed our first customer just last week, really. So we have already essentially this co-pilot project that we’re doing with a large French bank, and they’re interested basically in consuming the signal and essentially understanding how to use it in some of their systematic strategies. And yes, right now we have actually completed, I would say maybe a smaller benchmark, and we have already seen some very positive results. We’ve shared these results with other prospects, other advisors, and we’re getting some very good feedback. Basically, we’re seeing how these AI methods we’re using and we’ve already coded up actually can predict this five-minute signal, can predict essentially a five-minute out trend on the price, and they do it with some accuracy. So obviously there’s still a lot more that we need to do. So this is just our first benchmark. It’s kind of like a baseline. We’re already seeing some positive results and also the market is, again, kind of the people we’re talking to, is reacting in a positive way.
Yuval: Where do you see quantum helping in this at some point in the future?
Iordanis: Yeah, so there are quite a few different ways that we try to add what we have learned from all the quantum projects that we have done before, in particular the ones with the banks. And one way is to, once you have this AI pipeline and you have the model that takes your input and tries to predict the output, the question is, can you find the right place and go and put inside this architecture some quantum part and make it into a hybrid architecture where some parts will be done on GPUs, some parts could be run on a QPU, some coming back to the GPU and giving you the hard-earned answer at the end. So what we are trying to do is come up with some idea of a quantum agent, as I said, because in the end you want to play this game of trading. And we have some very recent results that show that indeed these quantum agents can be quite powerful, like they can discover new strategies or they can play games in a good way. But as we said, we are starting with a fully classical solution. We are ready to add quantum when the hardware is ready, when the algorithms are ready. And in parallel to building this classical MVP right now, we are working on improving the quantum algorithms that we want to run, at the same time hoping that the hardware will arrive as soon as possible.
Yuval: A few days ago I spoke with the CEO of Aqora and I think he told me that you guys are setting up some kind of a hackathon. Tell me about that. That sounds really unusual for a very early stage company. Is that true and what are the goals of this activity?
Iordanis: So yeah, we are collaborating with Aqora. What they wanted to do is come up, at least in the beginning, with some specific use case in finance and try to understand what quantum can offer as a solution to this use case. So we work together to provide a specific use case, including the dataset, including the benchmarks. And I think right now it is up on their website where anyone can go and provide their own quantum solution and see how well they do with respect to the action benchmark. And I guess the idea is that we start with one use case, but then we can add some more and there is a prospect of having a live competition, as you said, like a hackathon. But for now we have this use case on the website where everyone can propose a quantum solution and hopefully little by little we’ll get to better solutions.
Yianni: Yeah, and maybe to add to that, the objective is really to meet new people. And this is something we’ve done in the past also where we posted some interesting problem. I think this was popularized by Google some years back where you had to kind of solve the problem and fill out the rest of the URL and get to a page where you can submit your information. And we’ve seen that, so we’ve hired people like that in the past. And yeah, so actually the prize for the hackathon is a potential internship with Quantum Signals. So it’s just a nice way to meet new capable people and get to work with them.
Yuval: What is the experience in raising money for a quantum or quantum-inspired company these days?
Yianni: Yeah, good point. So I think there was definitely a time, I want to say maybe a year or so ago when things were harder. What we saw now over the maybe two or three months that we were trying to raise our pre-seed round is that, and since then, since then we’ve also had several inquiries. And there is definitely some interest. Now I think we’re also, the fact that obviously we’re an application company and the fact that we have this specific use case in mind in the specific industry, finance, and we’re using AI. All these are sort of qualifiers for my answer, I guess. But in the end we were backed by a quantum VC. So we were backed by Quantonation, which took over basically the entire pre-seed round. And we’d like to thank them very much for that. And so in the end we basically spent maybe close to two months going through these motions and then very quickly essentially getting to some kind of term sheet and some kind of agreement. So I think for quantum plus AI, there is definitely, it seems like there is a bright future. Now you could argue whether it’s quantum or AI that’s doing a lot of the heavy lifting there. But I think there is still definitely interest from investors in quantum computing companies.
Yuval: I assume that’s not going to be the only use case that you’re addressing. Can you share some of the things once you get large order execution done? What’s going to happen next?
Yianni: No, but this is in fact the strategy, to concentrate on a very, very specific use case. So I think what we’ve done earlier in our careers is go after many different use cases. And I think that was the right thing to do at the time, to go after different use cases in different industries. But I think right now it’s more important to really go deep in a specific use case. And really solve that, make sure there’s traction in the market, make sure the idea can be commercialized and the software can be commercialized. And then we can try doing essentially wider things. I think this is also common advice from VCs. And I even read it I think somewhere on Y Combinator’s website. It’s essentially to drive some kind of wedge. So start with a very, very, very focused, very specific use case and then go wide. And I think this is the case for application developers and software companies. I think obviously things could be different for middleware companies, companies that are building maybe dev tools or obviously hardware companies. But for software and specifically application software, I think that’s the advice.
Yuval: When you started the company, you surely looked at other use cases in other industries. Quantum is often connected with chemistry or material science. I mean, why finance or why this use case? What was the criteria that you used to select this particular one?
Yianni: Yeah, maybe I’ll start and then Iordanis can also talk about the specifics. So it was really driven by our experience over the last four or five years. So we had several collaborations with very large banks in the last four or five years that we worked together with Iordanis. And so we had already created a very strong network of partners and people we had worked with and collaborated and we trusted and they trusted us. And yes, we looked at several finance use cases and we did consider in the beginning, we did consider potentially other use cases in finance. But we decided to concentrate on this one. As I said, it’s still considered an open research problem, one of the holy grail problems in finance. And it is connected to many other use cases in finance as well. So this is kind of the reason why we went for finance and kind of why we went for this use case specifically.
Iordanis: Yeah, it’s also a very kind of complex mathematical problem in the end finance. So it’s quite interesting to try and try to crack it and start by seeing what AI can do with it and then what quantum can do with it. And it’s also the case that if when we, you know, as Yianni said, we were discussing all these questions with our clients in the last four or five years. What we saw is that the state of adoption of AI in finance is still in the beginnings. So there are a lot of things to be done even with AI. So this is why I think it’s a very good opportunity to start working on finance. And I, you know, it’s a more personal thing. I also think that, you know, I found people in finance much more maybe open-minded to adopt new technologies, whether this is AI or quantum than some of the more traditional scientists that work in other domains. But I guess that’s only me.
Yuval: Earlier in the conversation, you referred to large order execution as almost game theory. You almost don’t want to tell the other side that you’re interested in buying a lot or selling a lot. So is there a reverse algorithm? Is there an application for your algorithm to detect when someone is trying to execute a large order?
Iordanis: Yes, in the sense that what you can try to do is kind of predict what would be the next orders that are coming. And the information that you get from that is by looking at what has happened before. And where you see some sort of pattern of, let’s say, sell orders, then basically you can try and predict that, okay, these sell orders are coming because they’re part of a much bigger sell order, which means that the next orders will also be on the sell side and not on the buy side. So there are correlations that you can try to use in order to predict what would be the next big orders that are happening. So all of these kind of predictions are what will go into some sort of reinforcement learning strategy that will tell you in the end what is the right action for you to take.
Yuval: How would people pay for this? Do you envision this as a subscription, you know, use this algorithm and pay us a fixed fee, or are you making money when the algorithm suggests something better than a sort of simple strategy? How are you pricing this?
Yianni: Yeah, so initially we’re going to go with standard B2B enterprise software licensing practices, which is going to be a standard fee, standard subscription, possibly annual, and that’s the fee. And then you use it basically probably as much as you want. Maybe there is a case to be made for a per seat fee, but that also probably gets complicated. Yeah, so we’re not thinking of doing anything related to usage right now, or maybe even the size of the orders being executed, or maybe even the size of the potential benefit. I think initially that’s going to complicate things too much. So initially we want to just make customers happy and they can get a lot of value out of the software and we’ll just have probably some kind of flat fee, yes.
Yuval: As we get close to the end of our conversation, I’m curious, you both have been in the quantum industry for a long time. I think you’ve probably met the first qubit in person, and now you’ve spent several years in the quantum industry. What have you learned in the last, say, six to 12 months that you didn’t know prior?
Yianni: Wow, okay. I think obviously people maybe are expecting more now from quantum companies. I think there was a time maybe two, three years ago when everybody thought that the path to commercialization is going to be easier. And I think there was definitely a time when there was a lot of interest from industry. I think now quantum obviously also has to compete with AI methods that deliver essentially more readily available advantages. And so, yeah, obviously on the one side, the quantum industry still has this great potential and still needs to develop that and we still need to do the research. And I think it’s also different between different areas of the stack. How a hardware or quantum computing company has to deal essentially with all the ups and downs, how a middleware company deals with that, how an application or software company deals with that. But I think also now what we found out, again, over the last maybe 12 months or so is that the quantum computing companies, they need to have essentially a backup plan in case commercialization does not arrive in the next whatever the estimate is. So you need to have a plan of what happens in case the estimate is extended by another year or two years or three years or you pick the number. So I think that’s really the key point.
Iordanis: On my side, I think the one thing that I learned this last year is to be more optimistic in the sense that I see that there is a lot of serious work happening in the quantum field. So the quantum hardware is advancing a lot, error correction side, fault tolerance is advancing a lot, quantum algorithms are advancing a lot. I think there is a bigger community now that does some very serious work and this makes me more optimistic that it will work out in the end. And at the same time, I became more pragmatic in the sense that, as I said, I do believe that the quantum revolution will happen, but it will take the time that it needs. So it may not be next year or two years, but I think it’s a very exciting time. We need to put in the work, otherwise it won’t happen. So I’m optimistic it will happen and pragmatic that it will happen when it’s the right time.
Yuval: I think that’s attributed to Bill Gates. Revolutions take longer than you think, but then they’re larger than you expect. And maybe that’s going to be the case in quantum. So speaking of sort of greats, I wanted to ask you both a hypothetical. If you could have dinner with one of the quantum or I guess one of the finance greats, dead or alive, who would that be?
Iordanis: I think I had dinner with many of the people who are still alive, but I want to do so. I might have to go earlier in time. I think when it comes to quantum mechanics, what interests me more and more, it’s also kind of relation to other fields like philosophy and theory and how quantum mechanics and the way of looking at the world in a different way has affected both philosophy, even in art. And when it comes to this more philosophical kind of questions, Niels Bohr is the person that had the right way of looking at things. So I would be really excited to talk more to him in this hypothetical.
Yuval: And Yianni, how about you?
Yianni: I mean, you could go with someone like, for example, Adam Smith, the father of economics and the author of Wealth of Nations. I guess that could be, that should be an interesting conversation to say the least. Yeah, maybe that’s what I’ll go with.
Yuval: Wonderful. Yianni, Iordanis, thank you so much for joining me today.
Yianni and Iordanis: Thank you so much. Thank you, Yuval.
To subscribe to the audio podcast, please Spotify here