Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Engineering Intelligence feat. Allora
Recorded: July 1, 2025 Duration: 0:27:49Space Recording
Short Summary
In a recent Twitter Spaces discussion, Tyagi and Mike from Allura Labs explored the innovative capabilities of the Allura Network, emphasizing its role as a predictive oracle in DeFi. They discussed the integration of AI in yield generation strategies, the launch of the Allura Agent Accelerator to support emerging projects, and the upcoming Mainnet launch, all pointing towards a significant growth trajectory in the crypto landscape.
Full Transcription
Thank you. Thank you. Hey Mike, GM, GM.
Yeah, I see that you dropped as a speaker.
I'm going to send the invite again.
We're just going to wait for a couple of more minutes,
as I see more people are joining in,
and then we'll kick it off.
Hey, GM.
Can you guys hear me?
Awesome. I see a thumbs up.
Really appreciate you guys putting this together.
Absolutely. Am I audible?
Yep. You're great.
Awesome. Awesome.
So I think we can start off with introductions.
I'll start with mine.
I'm Tyagi joining from the CapEx account.
I'm the co-founder of CapEx.
And today we have Allura joining us for this Twitter Spaces Engineering Intelligence.
We have had the pleasure to work with them previously,
integrated one of their models into one of our ecosystem projects that's called Candy.
And it has been quite amazing so far.
We have seen a lot of updates go within Candy as well as CapEx.
So, Mike, I think everyone here is excited and would love to know more about you, your journey, and more about Allura as well.
Yeah, for sure. Excited to be here. Thanks so much for having me and sending this all together.
But just to give everyone a little bit of background on myself, my name is Mike.
I am the head of go-to-market over at Allure Labs.
I've been over here for about a year and a half now.
Prior to that, I was working at Chainlink Labs for three years and led the strategic business development team over there.
I have been in crypto non-professionally just as a hobbyist since I'd say like spring of 2017 is when I first got involved.
Amazing. And that's been quite some time now. I think we've seen a few cycles.
Yeah. Yeah. It gets tougher every cycle, I would say. Back in 2017, it was a lot easier to make
money. Just you really couldn't go wrong. Getting to an ICO and as long as you didn't hold on to it for dear life,
yeah, you would do pretty well.
And then the next cycle was yield farming.
And this time, I guess it was meme coins,
but it feels like it gets tougher and tougher to get that alpha as time goes on.
Absolutely.
I think the wild, wild west of crypto has finally kind of settled.
It's not as wild anymore as it used to be, of course. Yeah, it used to be a dark room and it
was easy to feel really bright. Now the room is pretty illuminated. You're competing against a
lot of other smart people.
Yeah, fair enough.
So yeah, one of the very first questions would be that, you know, what is Allura Network?
And how is it different from, you know, or how would you essentially put yourself categorically in the AIX crypto landscape. Yeah, so what drew me over to work at Allura from Chainlink is
what Allura is doing, there's nothing really else that's quite like it.
There might be some stuff that's like tangentially like, you know,
it's kind of like that, but mixed with this.
Maybe like a good way to think about it is as a predictive oracle.
So something like Chainlink,
if you want the price of Bitcoin USD or ETHT USD,
and you want what that price is right now,
you'd use Chainlink, you'd use PIF, something like that.
But if you wanted what the price of Bitcoin would be 10 minutes from now, an hour now, a day, a week, you'd use Allura for that.
So Allura, you can think of it as a self-improving decentralized collective intelligence network for the financial and behavioral applications of AI.
So it's not just price predictions.
It can be any sort of really financial metric, it can be volatility, implied volatility, how much liquidity, volume or fees are going to be in some
liquidity pool over some period of time, APY for things like LSTs, LRTs, or just some DeFi
yield strategy, interest rates for certain assets on certain lending protocols. And then you start
to have this crystal ball, this predictive primitive
that you can use to start powering much more sophisticated and much more dynamic strategies
than what you've seen historically and still see today within DeFi. Like what you see within DeFi
today and ever since kind of the beginning is it's very static strategies. They're based around linear math. They're reactive to actions as they happen.
But when you're able to have a good sense of alpha
and know that scientifically there is alpha
and a level of predictiveness embedded
in this inference that I'm receiving,
you're able to do a lot more interesting,
a lot more expressive things with that data.
So that's what we're helping enable.
Hopefully that gives you guys a little bit of background
on what we're building over here.
Absolutely.
And this also brings me to the next question,
which is you talked about all of these prediction
model that you've built out.
And I think that's also one of the things that we integrated.
How do you see, like, what use case do you see as the most prominent one when it comes
to deploying Allura into the market?
And specifically, let's say, within this AIX crypto landscape, like which companies or which products do you see leveraging Allura and for what use case?
So, I mean, I can give you like the broader answer and really like the broader answer is, I'll give you a more specific one too, but the broader answer would be basically almost
anything. It's like 80 to 90% of things that you can think of today in DeFi that happen,
those can be improved by AI in some aspect, whether it's user experience, capital efficiency,
automation, a lot of stuff.
There is room for improvement, some more so than others,
but almost everything can be improved
by having a predictive inference inserted in it somewhere.
I'd say the ones that are earlier on where these are the things
that we're getting interested in here and now,
and teams are actively developing them,
or some of them are actually live in the wild.
Agents are one of the biggest narratives
that we have in Web3 today.
A lot of teams are building agents
that are focused on yield generation,
and that's really what we're trying to help enable
in this first wave.
Stuff like risk management and risk parameters,
that's also being worked on,
but the stuff that's like yield generation,
I think that's gonna be the very first wave.
And we've already seen stuff go out and be very successful.
But things like, I like to call it a spectrum,
where you have your more conservative strategies on one end
and you have your more DGN stuff on the other.
Yeah, kind of walking from the conservative end to the DGN end.
I'd say like intelligent DCA strategies would be, you know,
one of the most conservative things that you can do with Allura
or like building a predictive strategy.
So instead of just saying, hey, I'm going to buy $100 worth of Bitcoin every hour for the next month, what you can do with Allura, because it has these
predictive inferences, you can scale appropriately based on if that price was forecasted to go up or
down. So let's say you wanted to buy it every hour for the next month. All right, one hour from now,
two hours from now, three hours from now, it's predicted to go up. So you actually want to buy it every hour for the next month. All right. One hour from now, two hours from now, three hours from now, it's predicted to go up. So you actually want to buy more than you typically
would during that epoch because the price is going to be going up into the future. And conversely,
if the price is forecasted to go down, you would scale that buy and buy less than you typically
would at epoch because you think you're going to get a better price in a future epoch. And
naturally, if those predictions end up being over 50% correct,
then you should come out with a better cost basis
than you otherwise would with a traditional DCA strategy.
Then kind of like going more towards like the middle of that spectrum,
I'd say things like automated looping strategies.
So instead of just doing a rehypothecated loop where you set it up with a one-click deploy,
but then once it's instantiated, you need to then de-risk yourself if your collateral
asset starts losing in value.
You can have AI take a forecast for price and volatility of that collateral asset, increase
leverage when things are looking good, decrease leverage or move you entirely over a spot when market conditions are looking a bit bearish. And then as we start
pushing further and further along on that degen curve, you have things like delta neutral trading,
directional trading, directional trading with leverage. And we're seeing teams that are
actually doing directional trading with leverage and actually doing it rather well.
So I'll just give a shout out to a couple of teams real quick.
You have Vectis and Robonet.
Both of them are hosted on Driftrade on Solana, but they are trading sole perps with leverage.
I think Robonet goes up to two or two and a half leverage.
And then Vectis, I believe, goes up to 4x leverage. I think Robonet goes up to two or two and a half leverage. And then Vectis,
I believe, goes up to 4X leverage. Both of them have been live for, I think, Vectis around
two, two and a half months. Robonet, maybe around three months or so.
Robonet changed their strategy recently, but it's done like a 14% APY increase over the past, I think, three weeks since they updated their strategy.
And then Vectis, that's been live for a bit over two months.
And it hovers somewhere around between like 13% to 20% annualized APY.
So we have teams that are doing the most decent thing that I mentioned on that spectrum.
And they've been live for a pretty decent amount of time.
And they've been trading with leverage successfully.
So I think that just goes to show you how powerful this is.
And really looking forward to seeing
what other teams in this space and just people that are building
can really develop with these predictive inferences.
Really, there's not much limitation.
It leaves a lot of room for creativity and expressiveness.
Amazing.
And in order to, like, you know, get to this stage,
could you also share some insights as to what did, like,
how did you guys approach this problem?
Because, you know, intelligence in itself,
and especially when it comes to Web3 and markets,
I always feel like, you know, intelligence is, intelligence always has that sort of a double-edged sword where in order to, like, give it, give intelligence complete autonomy,
it's something which we are not there yet, you know, because there are always chances
of it hallucinating, there are always chances of it, you know, taking decisions which might not be in the
best interest of the trader.
So how do you guys overcome those challenges when, you know, building this sort of a stack?
Yeah, absolutely.
And I think like one of the more common misconceptions, and this is like where I found myself when
I was just starting to learn what they're building at Allura is
everyone kind of just thinks of LLMs. When you think of AI these days, like most people,
they think of LLMs, like chat GPT, maybe you think of image generators or text-to-speech,
stuff like that. But really, the OG, the granddaddy of AI use cases has been financial
applications. AI has been used in banking and trading for 30 plus years.
It's a much more mature part of the industry.
And that is more of an objective centric approach where you have these narrowly defined models.
They have a very tightly scoped objective function within them.
And that's how the lower network is set up is
you have a defined objective as the topic. So like Bitcoin has ETH USD price feeds,
we would have something like a Bitcoin 10 minute price feed. So like, what will the price of Bitcoin
be in 10 minutes? And then you have people that you have two sets of what we call workers.
You have inference workers and forecasters. The inference workers, those are people that are
running disparate machine learning models that are purpose built for that defined objective
of the topic. So, you know, it can be an open source or closed source model. They can run this
model anywhere that they want. They don't host it on the network itself. It can be on AWS or GCP.
It can be on bare metal in their mom's basement. They're totally agnostic from that point of view.
All that these inference workers are doing are taking the outputs from their models and then
submitting it to the network. So that's the inference worker part. The other end of the
equation is the forecasters. The forecasters, they have contextual awareness of
what the current market conditions are, as well as the history of which models in that topic
performed the best under similar market conditions in the past. And just by the nature of models
themselves, some models are going to perform best in highly volatile markets, some in sideways
markets, some in low volume, some in high volume, et cetera. So those forecasters, they know which
of those models perform the best under which market conditions,
where are the current market conditions now,
and then they use that to run what's called a loss function against all the inferences that they received during that round
to basically predict which of those predictions is going to be the most accurate based on their historical accuracy
and their similar market conditions in the past.
So then they aggregate all those together.
The result is just a single weighted output.
And the whole kind of like theory
behind what we're building over here,
and this is what we see in our research and testing as well,
is the output of the network consistently outperforms
any of the individual models within it.
So that's where you get this collective intelligence.
And why this is important is LLMs,
why they're great at many things
like predicting what the next word
in a sentence is gonna be.
One of the things they're actually very, very poor at
is doing any sort of financial modeling
or financial forecasting or time series predictions.
So, and this is what we're running into a lot of teams
that we're speaking with.
You had this big agent boom around like november of last year and you know teams have all
these great ideas uh but then they think they can just use gpt for running some sort of trading
strategy and then they start doing the testing and they realize hey this training strategy is
not very good it's losing money very very rapidly uh rapidly. We're not sure how to solve this.
And yeah, it's because you have LLMs
and those LLMs just fundamentally aren't good
at trying to do like these time series forecasts
that are necessary when you're trying to have
some sort of yield generation strategy.
So what Allura does is it allows them to use something
that's purpose-built for these sort of financial metrics that LLMs are bad at.
And it's not replacing the LLM.
You would just typically make an API call to the Allura network as part of your GPT wrapper.
Allura network gets back and says, okay, the price of Bitcoin in one hour is going to be this. And then GPT takes that, sort of extracts away that part
that GPT is bad at,
takes it from Allura,
goes about the rest of its flow.
And now you actually have something
that you can build into an agent
that can do some sort of
yield generation for you
and be able to do these
sort of financial applications.
Got it. This sounds very exciting. I mean, I remember you dropping a couple of projects which are already leveraging Allura. Is there anything which, you know, probably something that everyone who's joined this
space can take away and probably like, you know, try out how it actually works?
Maybe some sort of a live demo or maybe, you know, maybe you guys have some sort of
playground where somebody can go and go and get how it actually works. Maybe some sort of a live demo,
or maybe you guys have some sort of playground
where somebody can go and check out
how these predictive modeling works.
Yeah, so if you want to go into a vault,
see this is like kind of the tough part
about that question is we're back-end infrastructure.
So it's like, hey, go use Chainlink. It's like,
okay, well, go use Aave, go use GMX, go use any of the protocols that use it. But yeah,
it's not something that you touch like that. It's on the back-end at the smart contract level.
But yeah, I would say if you go to Driftrade and you go to their vaults, you can see one by
RoboNet Finance that trades SolPerps. That one's been doing rather
well the past few weeks. And then if you go to Vectis Finance, they had an Allura-powered,
I think it's called the Allura AI Edge Vault. Both of those are trading SolPerps on Driftrade,
and they've been doing it pretty well. So you can see how the vault's been performing.
I think on Vectis, you're actually able to see a chart of all the buys and sells overlaid on top of a price chart.
So you can see when it's buying and selling and how the prices reacted afterwards.
And, yeah, it's probably a good way to just start getting your feet wet, checking it out.
But again, we're back in infrastructure
that teams are using to build some very cool,
some very interesting strategies
and realize the next generation of DeFi.
You're going to see more and more protocols
start incorporating AI into things like,
I'd say these agents and yield generating strategies
are probably going to be
like that first phase. Then you're going to have some like let's say optimization things like
dynamic fees, liquidity provisioning. Those are going to be AI enabled. There's already
liquidity provisioning that is AI enabled and we have a case study that we did with Steer Finance on the results of that. And they actually blew away what our best expectations were going to be.
So if you guys have a chance to check that out, would definitely recommend it.
And then I'd say kind of like the third phase would be sort of risk management stuff for lending protocols, that's probably a little bit longer tail out.
But what you'll see, I'd say 12 to 18 months from now, is it's just the norm for AI to be incorporated within most aspects of any sort of DeFi protocol.
It's just inevitable. It's a much better developer experience. It's a much better user
experience. The capital efficiency is a lot better than what it was. So in my mind, there's no way
that it doesn't happen. We can argue about timeframes and depth of the penetration, but
it's going to happen. Things just naturally improve in the universe. And this is definitely a pretty big
leap forward. I think we're still though, maybe not even in the first inning where people are
just starting to hand in their tickets and file into the seats. But give it some time and you'll
see that the majority of on-chain transactions are going to be through either
agents or using some sort of AI interface. It's going to be that meme like, back in my day,
we used to have to go and put loans on Aave ourselves and click to approve every single
transaction. And you have the grandkids saying, all right, grandpa, let's get you to bed.
I just replay that in my mind. I'm like, this is where it's going to be going. And it's going to happen a lot quicker
than people really realize. I think I absolutely agree to that. I mean, already everything that
we used to think that it's going to happen five years down the line has happened in the last six months so we're definitely moving in that
direction um this just brings me to the last question i know this uh this is this has been
amazing spaces so far um you have been super sharp and i think everywhere everyone here uh learned so
much about allura and you know what you guys have been building um i came across this cohort of Allura Agent Accelerator and this
was this got me curious as to you know what all what what is all this about and I think as also
as part of like you know something which comes under your purview of responsibility I think as
part of GTM this this would be something which which is like very interesting so if you could
like you know share some words about what Allura Agent Accelerator is all about
and how can somebody who's building in this space
could become part of it.
Yeah, so anyone that's building in this space,
whether it's an agent or some AI-enabled use case
that they have Allura in mind for,
we're going to be setting up an accelerator.
So a little bit of a grant and incubator program.
If you're familiar with like SaneLink and what they did with Build, it'll be a bit akin to that. So that's set up
primarily for teams that are, I wouldn't say even like pre-series A, I'd say, but like younger
teams that are developing, can lean heavily on AI as part of their go-to-market and see
Elora as one of their key infrastructure partners and something that they rely on.
If it's a new, interesting use case that we think has a lot of potential and people are
wanting to develop on it, we can lend resources to them, whether that's through a grant,
whether that's through expertise,
whether that's through business development and networking.
We can help these teams really accelerate what they're doing.
That's what we want to do is we want to help accelerate this space
as quickly as possible,
making it easy for teams to connect with who they need to connect with.
You know, they may be great developers, but they haven't worked in the industry and have those
connections just yet. So they need someone to open doors, make introductions for them. Maybe they need
help funding on the grant side. You know, we can help incubate these teams to level them up,
to get them to where they need to be so they can actually be productive companies that are building really cool stuff that is going to be staples of the industry moving forward.
Awesome.
I think that's also very much similar to what even we have been trying to do.
been trying to do. But I think this is again a great value add specifically for projects
who are trying to solve for the DeFi space and or maybe just projects who have a lot
of use case when it comes to prediction AI models and can leverage Elora for that. So
kudos to that. And any ending parting notes, Mike, that you want to share with the community,
please feel free. We are almost towards like have come to the end of this spaces.
Yeah, just wanted to thank you for having me on.
I see a lot of people in the audience.
Hope you guys learned something or at least was able to take a break from your normal
workday and enjoy the conversation that we're having.
We are going to be launching Mainnet rather soon, working around the clock to make that
a reality. We have new Forge topics coming up. So if there's any model developers in the crowd
and you're looking for a way to get involved with LoRa, we do these Forge competitions that are a
little bit like, you know, call them model thons, a little twist on a hackathon. Or if you're
familiar with like Kaggle competitions, it's a little bit of
our spin on that. So you'd be competing with others on building out the best, most accurate
models for typically financial metrics. We have some probably around eight to 10 right now that
are predicting various blue chip crypto assets for a seven day time horizon. So if that sounds like something you want to
participate in, we'd love to have you. Other than that, if there's anyone that's working on an agent
or any sort of a DeFi protocol where you think AI can help enable a new cool use case, my DMs are
always open. Feel free to shoot me a message. Very friendly. We'd love to have a chat with you.
feel free to shoot me a message.
Very friendly.
We'd love to have a chat with you.
Awesome.
Thank you so much, Mike.
This is an amazing spaces.
I think we should do more of these.
I'm sure this was a value add to everyone who joined in as well.
Specifically, if you're someone who's building in this space,
just hit up Mike or join the Allura ecosystem,
join the Discord, follow them on Twitter,
be out there, take all the chances.
There's so much opportunity in this space coming up.
If there's anyone who's building in this space or
just curious and wants to just hang out and
brainstorm on some ideas, I'm always available.
My username on Twitter and Telegram is Tyagi CapEx.
Feel free to reach out.
Always there to help you guys out
and make the necessary connects.
So yeah, cheers.
Have a nice one, everyone.
Thanks so much for having me.
Have a great day, everyone.
