# BBW25 - Day 3 - AI, Data & Market Intelligence: Extracting Signal from the Noise

- Channel: [Boston Blockchain Week](https://streameth.org/boston-blockchain-week)
- Date: 2025-10-07
- Duration: 27:00
- Watch: https://streameth.org/watch/yt-XHLbfLckg2Y
- YouTube: https://www.youtube.com/watch?v=XHLbfLckg2Y

## Description

Boston Blockchain Week 2025 - Day 3
Thursday, September 11, 2025

Panel - AI, Data & Market Intelligence: Extracting Signal from the Noise

As digital‑asset markets mature, the volume and complexity of available data continues to grow. In this session, leaders will discuss how they combine on‑chain and off‑chain datasets with machine‑learning techniques to uncover trends, detect anomalies and generate actionable intelligence. From predictive models that inform trading strategies to dashboards that surface network health for builders, this conversation will highlight the tools and approaches turning raw data into insight for traders, researchers and communities.

• Austin Bennett, Research Software Engineer, Circle
• Kyle Fugere, SVP Engineering, Flipside Crypto
• Heidi Pickett, Chief Business Officer, The Tie (moderator)

## Transcript

Appreciate it. Thanks. Appreciate it. >> Okay guys, this is day three. We have two more panels left to the day. I thank you for your uh patience and for sitting here and enjoying this program. Uh the next panel that we have is on AI data and market intelligence extracting signal from the noise. Uh joining us on stage is Austin Bennett who's a research software engineer with Circle, Kyle Fuer, the SVP of engineering with Flipside Crypto. And this discussion is going to be led by Heidi Picket, the chief business officer at the Thai. Please join me in welcoming them to the stage. [Applause] Thank you, Ian. I am very happy to be moderating this panel. Super important topic. And we're almost the last panel, which would I would have said they saved the best for last, but I can't say that because there's the one more panel after us. But in any event, I'm Heidi Picket. I am chief business officer with the Thai. If you don't know us, we are a uh provider of information services for institutions and digital assets. Super excited to be joined here uh with Austin and Kyle from organizations that probably don't need any introduction, but we will get to that. Um on what I think is probably one of the most talked about topics today, and that's the intersection of AI, data, market intelligence, can we extract uh signal from the noise? So excited to be here with you. Thank you so much. >> Yeah. No, we're excited. >> Okay. Well, let's kick it off. Everybody knows Circle. Everybody knows Flip Side Crypto, but let's have a quick introduction. Who you are and more specifically, what is your role at the intersection of AI and data? >> I'll go first. Thanks, Austin. Um, uh, name's Kyle Fuer. I run engineering and data at Flip Side Crypto. um been at the company for three and a half years and really helped kind of push our our effort into the AI space. We've always been in the data space. Uh the next big evolution for us was how do you combine kind of AI and the rich data asset we have and so uh looking forward to speaking about a lot of that here today. >> Hi everyone, my name is Austin. Uh I joined Circle four months ago actually on the research team where I primarily cover AI. Before that, I did my uh bachelor's and masters in statistics AI at Stanford uh and helped create the stable point class there. So, >> nice. So, I would say like it wasn't that long ago, maybe a year or so, but like it was all like, okay, AI, crypto, is it just hype or is there something real? So, let's just kick it off like I think we've moved beyond that narrative and that it's much more positive. there is something real here. But I don't know what are the myths out there that you want to just say that is not accurate. Let's move beyond. This is like something we need to be all paying attention to. >> I feel like you can take that in so many different ways, you know, like um is AI a fad or is AI magic or even the combination of AI and blockchain could be um you know is that for payments? Is would we see microp payments in the future? But um you know I'm a bit of an optimist on this on this space and so I think the biggest myth currently is that AI can't be trusted for a lot of this um data analytics or analysis. Um you know we we certainly had a period where there was a lot of hallucinations and um people kind of built that into their heads but I think increasingly we rarely run into hallucinations. What we're running into is, you know, it just searched the wrong table or it didn't have the context to answer the question. Um, you know, there's a lot of like, uh, I think of autonomous vehicles in this way. You know, they're far safer than human drivers, but when one crashes, it's a big deal. And when your AI messes up a little bit, it's a big deal. Um, even though it's probably more accurate than most humans. And so I think the myth is that it's not yet at the stage where it can do largecale data analysis uh effectively. >> For me I think it's just thinking of AI is a monolith. Uh like whenever I talk to people they always think like oh AI is chat GBT when in reality AI is supervised learning like classic statistical modeling or deep learning unsupervised learning. There's a lot of different things it is. So, thinking that you can kind of have this one-sizefits-all approach to AI with blockchain or anything really, I don't think that's valid. It's it's a myth that a lot of people have. >> So, I mean, Kyle, you touched upon this a little bit, but I guess the next question I have is around just the sheer amount of data and it's growing like all the time, particularly in uh digital assets. So, what are the biggest challenges that you're encountering today? Is there like too much noise, lack of standardization? Um, you know, what's the difficulty in extracting like real insights, actionable insights? >> I'll take I'll take this one, too. Um, so the it seems like a boring answer, but we we have 40 trillion plus rows of data uh currently. um that's going to be 80 trillion rows by next year if Stripe and others fully integrate uh stable coin payments across every transaction. Um you know I I'm already worried about how we store and access that data. Um we're already having those conversations internally of like what is the appropriate amount of data to store? Um there's a history here. Where does it live? Um you know certainly we have the aggregates and what you know you see and search within our system is is the aggregates of that data um but that raw data happens and it lives somewhere and so as I think about uh the adoption which is amazing um you know for for many of these chains that that data is just massive massive volumes and and so for me it's it's really the the storage problem as much as is the uh analytics problem we can we can curate and model data effectively to make it searchable. Um, but that raw data still lives somewhere and so that's that's always on my mind especially as the owner of our our budget on the uh data warehousing side. Um, but that's that's one of them. >> Yeah. Uh, of the three you mentioned and just as it applies to like at least my work specifically, it's probably noise. Um, because generally one of the great things about blockchain is data is super auditable. I mean there's all these transactions for anyone to go view. Uh so like we regularly audit stuff like it's it's generally quite available but noise can be really hard to sift through if you're looking for specific insights. So like one example is uh I think something that I see a lot or like hear a lot is like oh crypto is just driven by like gold like they're highly correlated when like increasingly that's not that's actually becoming not true. um for stable coins it's very much not true and so it you know like you'll see like oh there's a moderate correlation between like so I'll use USDC as an example like between uh USDC and the gold price like oh so that must be there but in reality like a lot of that is just noise if you look at where USDC is actually correlated with it's uh deposits uh M2 it's actually it's far more correlated so like I think with crypto because we kind of all see it moving in one direction and like there's so much weird interaction between everything people lump it all together and it makes it really hard for us to sort out, hey, what's actually driving each of these things? So, >> um, so it's really interesting about the correlation with gold or or lack thereof, but what other insights are are you able to glean just from the sheer amount of, um, data around USDC that you have access to? >> Oh, uh, well, like that's, you know, publicly available stuff. Uh there's a lot of very interesting differences uh with USDC even between USDC and Tether like how they're transacted um what the distribution of transaction sizes looks like as you'd think it'd be well if you remove Tron from Tether it actually is quite similar uh but uh you know there's a lot of differences and so those are some uh for USDC I think just as someone who kind of joined circle and started looking at this information um I was surprised just how much volume there really is with USDC like I I did not think it was being as transacted as it is. Like USDC especially as of recently has seen like this explosion in transaction volume. Um and it's not just like a bunch of whales moving money around. Like it's like actual EOA wallets. It's pretty interesting. >> So what's driving that? >> Huh? >> What's driving that? >> Uh just if you were to ask me, I think it' probably be USDC adoption, but that's pure speculation. Just pure we're seeing more wallets, we're seeing more transactions. So that that would be my guess. >> And and back to you Kyle. So you talk about you know one of the issues just around you know storage of this sheer amount of data that's going to be like doubling in size in one year and gosh who knows like you know how that will continue. What other uh challenges do you face in terms of particularly around you know scrubbing the data and making it into something that's actionable but like in a timely manner like that's there's just like a lot going on like you got to have good data but you also have to have timely data or you going to or somebody's going to miss something >> 100%. And so you know I I always try to differentiate there's there's real-time data and then there's historical data. Um, and so we we tend to work in the historical data space. And if it's an hour of latency, that's historical data, right? Um, you're not playing in the real-time space. Um, and in the historical data space, you know, luckily we have we have seven years of trying to work on this. Um, you know, Flip Side's been around for a while. We've always been on the data side. Um, and there's been a lot of work and iteration to really try to standardize as much as possible across different ecosystems. And so you have your EVMs, you have your SPMs, and you have your oddballs. Um they they all try different things. Your move chains. Um there's different ways of interrogating and modeling that data. Um and so I think the bigger challenge, especially as we bring in a new partner or a new chain, is um spending the time to really understand the mechanics of the chain. And if we can do that, then we can model it. Um you know, we've gotten really good at onboarding. you know, at this point, an EVM is a flick of a switch, right? Um that that's so standard. Um but we still work with a lot of U partners who are trying something new and there will be more, right? There's going to be new innovations on on the tech stack and and so we just have to stay ahead of it. Luckily, we have a pretty amazing uh data engineering team um who who spends all day and night kind of thinking about these problems and and perfecting that. Um and and I would say last point is, you know, we're not afraid to invest in the right tooling. Um whether that's AI tooling or, you know, all of our data is in Snowflake. You know, let's make it easy on ourselves. We don't need to reinvent the wheel. And so that's that's really how we approach these problems. >> Yeah. Yeah, I don't know if you have anything to touch upon that in terms of um you I assume you don't have to really scrub your data like like flip side would need to given it's just coming through your own you know USDC and such but is there any issues around you know how do you get those actionable insights quickly as opposed to like sorting through? >> Sure. Yeah. So that's I was say that'd be a better question for our data engineers. Um but on research for what uh at least I do for trying to find insights just like standard process of exploratory data analysis and then maybe some baseline models uh sometimes cleaning if I'm trying to get sources that's like not contained within our data um like for example I was trying to size uh like the amount of like transactions with USDC and some other token uh and I was looking at like dex volume and that's really hard to find is like the amount of dex volume or like trying to get an estimate for how much there is just generally >> um bringing um AI models into the discussion. Um so obviously only as good as the data they're trained on. How is Flip Side leveraging machine learning to improve the accuracy and usability? >> Great question. Um you know AI is really really good at writing SQL. It's really good at coming up with a plan. Um it's really good at leveraging tools. Uh I think where it struggles still is if it wrote a query that returns a large data set. Um can it effectively uh analyze that data set? Right? Did we overwhelm the the context window of that of that model? Um and so we've started to experiment with using uh really machine learning models or or classic um statistical models to kind of be an intermediary between the AI writing the query um interrogating the blockchain data and passing that through a statistical model and then asking the AI to interrogate the uh result set of that of that model which is a much smaller sample uh and actually creates a much better analysis of the data. you know that that may change these context windows continue to grow. Um but for now those classic models still work really really well for that type of scenario. >> Um so I just read a quote from um Mike Novagratz obviously CEO of Galaxy. He made a prediction. AI agents will become the largest users of stable coins transforming how everyday transactions are handled. So, Circle has made stable coins usable for businesses and institutions. How do you see that extending to AI agents? >> That is something we talk about all the time. Uh, so rest assured, we are definitely looking into this. Um, and yeah, I think that uh stable coins are actually a supernatural fit for AI agents because of the ability to conduct micro payments. Um, so like I think that you're probably going to see a lot more of that as we have agent interoperability, especially for agents that are like really really good for like one specific task, which I think is kind of how we're developing um, frankly. And so that's where you would see that. So I think you will see more development with that especially with like new protocols like I don't know if you guys have seen Coinbase's X42 open source like that's a good example for for payments um of something that I think could serve as like a baseline or something like where companies are starting to really go after that. I know circle we're looking at it actively. >> Yeah. Can I actually uh build on that? I prior to the flip side, I worked at a a retail tech um company uh and I ran their venture fund and in fact that's how I first got into blockchain, right? And the reason why we were looking at blockchain was uh the margins for most like grocery retailers, Walmart, Kroger, Tesco in the world are really slim and one of you could never do small payments. The transaction fees were too high, right? Um if you're if you're always facing a two 3% transaction fee um it really limits what you're able to do. Um stable coins and microp payments virtually eliminate that, right? It it is a huge unlock for um new business models but also new like pricing models there. There's a huge margin. you've pretty much doubled your margin uh if you can remove those fees for a lot of, you know, certainly grocery retailers in the world. And and to me, like that's always been the the huge unlock that we've been waiting for. Um when will this get mass adopted? Because you can't really take advantage of that. No one's really going to buy with uh no one's buying anything with Bitcoin, right? No one's buying anything with native tokens. Uh it's it's going to be stable coins. Um, and now that that's hit the mainstream, I think that truly truly um unlocks everything we've all been probably watching and waiting for. >> And how is um Thank you for for adding that. That's interesting. But also, but going back to the um AI agents a little bit too just in terms of how are you incorporating or flip side incorporating the use of AI agents in your business model? >> Uh well, there's there's two parts. you know, we we were early adopters to AI tooling, just running the engineering and data or um you know, the because we were such early adopters, I think it helped us actually build a good AI product, right? Our engineers knew how to use AI tools and knew what a good AI experience would be. um that was critical and you know I think the going back to the myths um you know the thought that AI will replace a technical user I just haven't seen evidence of you know certain roles for sure um but it's made our best engineers our best uh data analysts even better right they are the best users of this tool um and and so yeah definitely used internally Um but increasingly exciting for what the external product is like uh we have everything that's ever happened on chain right and now you can go and ask any question you ever dreamed of asking and our AI system will go and find the answer if you want to know what um were the most profitable wallets yesterday on Ethereum you can go find them and track them right if you want to interrogate an exploit that happened you can go type that in and it'll go give you a full on analysis of what happened. It's it's really uh very cool to watch and and that part does feel like magic when you first see it. You're like, "Wow." Uh how did it work through all of that? Um and I think it speaks to the progress we've seen uh with some of these more advanced models. I mean this seems like a major unlock the use of um AI to be able to just ask the questions and not having to have that necessarily technical background know how to use an explorer know how to like you know figure out and find that information like this is huge. It's huge and and you know for the first time like we we've had a SQL interface forever, right? Um but SQL doesn't give you the intent behind why they're writing that SQL. Um you know for this first time now we have the intent. Oh that's why you wanted that. Um and that intent is really really important and I think uh underappreciated perhaps. Um, it's it's really informing where our product's going because before we could interrogate the SQL. We knew what it was trying to do, but we didn't know the intent behind the the question. >> Yeah. >> Did you want to add something? >> Sure. Yeah, a little bit. I was going to say I as someone who a lot of my work functionally is just data science um for like strategy, all this sort of thing. Um, and it's incredibly helpful to literally be like, "Hey, can you go run cross validation on like all this stuff?" and like it can just do it in like 30 seconds as long as you prompt it well. Um, also I don't know if we have any developers who do it frequently, use cursor rules files. They are your best friend for this kind of thing. Um, but yeah, it's really interesting. The one like danger I feel like though with uh AI interpreting statistics is it's like they will kind of or at least I personally observed that they'll like jump to conclusions off of like correlation or something um and like completely ignore other issues like overfitting for example. Um or like they'll like have some weird balance of like all these things. Like I I think there does need to still be human in the loop a lot of the times if you're doing like actual like you know causal inference like stuff like that. But for like a good baseline like hey how are these two data sets related? It's really good at a first pass version of like oh they're like pretty strongly correlated. Um and it can do that like you can literally throw into like GBT like or five thinking. I know they they change it now. Um I use claude but uh and uh it will it will like actually generate like a summary. It can even generate like plots for you. It's pretty cool. So anyway good is it always admits when it's wrong. He's like, "You try again." You're absolutely right. >> Yeah. Oh, the we talk about the uh syncopy and and how DPO I think it's that, but I'll ignore that for now. >> Um, do we have any students in the room? >> Two students. Are you interested in AI and data and crypto and digital assets? Yeah. What What skill sets are needed for these two to get into the industry? What do they need to be learning in you're in school? >> Yeah. Yeah. Students in school. What do they what do they need to uh to to become one of you in in your roles? >> Okay. So, I just graduated so I can I can provide some good things. Um if you have not uh probability class at some point you need probability theory. Um very helpful just like every I think everyone should take that. Uh a data mining class is really good. Um you don't need all this fancy like to get into this kind of stuff. you don't really need to start with all the like super advanced AI whatever. Um you're better off taking like a classic like learning about regressions like what are the problems you see with that. Um and then then building on that then you can get more into like linear algebra all the stuff for like actual machine learning. Um I'd kind of recommend going in that order because honestly to do like a lot of complex AI like stuff you need to understand that basic statistics like you need to understand what are the signs of overfitting what is it look like when something is like biased how do I build these kind of things and like just taking classical like models normally they're called data mining at most universities um also if you want to do my area which is reinforcement learning stochastic processes if you want to see that so that's a that's a fun one too. It's like kind of how things move with time and time series. Sorry, >> it's been slightly longer since I graduated, but um you know, for me it's curiosity. You can literally learn anything now, right? Everything is there. It's accessible. Uh a new programming language, whatever you you can dive right in, you know. Uh it's funny, we we started to um to to write in languages that we never supported previously because we actually just found they were more efficient um at certain things like we we would always previously gravitate towards Typescript because everyone knew Typescript right now we can ship and go and that's okay that I don't have a ton of Go experts. Uh the AI can be the expert. And so uh curiosity is really the number one skill that I look for when hiring personally um because it is a strong indicator that you will leverage the tools at your disposal to uh build something, right? You're not in it just for the the art of the language. Um you're in it to build something. And and I think that's that's truly going to be the skill set that will set you apart. >> Fast forward one year. If Ian were to say, "Please come back and do this panel 2.0," what do you think we'll be talking about or not talking about? >> Well, I I think you touched on one, which is stable coin payments for agentic AI is a really big one, and I think it's just only going to get bigger. Um, there's probably going to be some new advances in LLMs, as there always are. Uh, I would think you probably see more um integration, not so much like actually like AI systems like being done like on blockchain so much as you kind of see them working in parallel. At least that's how I kind of see it. I think you're going to see a lot more of that soon. Whether the LA second the other one like of them actually being like together comes into play. I don't think in the immediate future, but I think you're going to see a lot more like hey, these two things are working together while being separate. So anyway, >> I think it would be we would be very interested to know the percentage of transactions that were initiated by a human versus initiated by an agent. Um I think increasingly I mean it's already happening right we call them bots a lot of times. Um but um you know I think we'll want to start to differentiate and draw a line and and truly understand that right are these are we growing healthy activity and is that healthy activity being driven by humans or agents and I think that truly will be kind of where we start to interrogate deeper certainly into blockchain data is is understanding that dynamic. Yeah, I was say I know Stripe did some interesting research on this like a gigantic model of trying to model like where payments are coming from and trying to cluster them using that. So >> do you think AI agents will take our place on this stage? >> You know I I think we will always value uh human time more than we will uh agent time and so I I hope not. But maybe it would be a nice mix to have a few agents alongside us. Oh, no. No. >> Uh, yeah. Just no. >> Um, any final words you're most excited about or uh I don't know, anything, anything you want. You get the, uh, the closing. >> Uh, I'm just excited to see the future. Like, I can tell you firsthand, I know a lot of people because I come from the AI side originally who are getting interested in blockchain and I am very excited to see more like people who are kind of like AI native become more into crypto. Um and yeah for any any people who are looking to study it like please do it' be fun. >> Yeah. No I'm just excited to see the progress you know it was uh wasn't the easiest four years here in the US environment like the the regulatory environment and so to see the innovation unleashed is really really exciting and um you know I think there's probably stuff that's going to emerge over this next year that we haven't thought of. Um and so I'm excited to see what that is. Amazing. Thank you. Thank you everyone for uh joining us. Thank you Ian for an amazing uh blockchain Boston blockchain week. Really appreciate your leadership and bringing everybody together. Um appreciate the time on the stage and um yeah, have a great rest of your day. [Applause]