EXCLUSIVE INTERVIEW - Inside HELIX: Kaiju Worldwide’s $800M AI Trading Strategy Explained

This interview was conducted in writing and has been lightly edited for clarity.

 Fresh off the announcement of an $800 million valuation for its AI-operated trading strategy, HELIX, Kaiju Worldwide is making headlines in the world of predictive artificial intelligence. At the center of it all is Ryan Pannell—founder, CEO, and Global Chair of Kaiju—whose unique background in theoretical physics, cryptography, and finance has shaped the company's bold approach to AI-powered investing. 

In this exclusive written Q&A, I reconnect with Ryan following our previous interview at Ai4 to dive deeper into HELIX, the future of predictive AI in financial markets, and how Kaiju is pushing boundaries through innovation, responsibility, and collaboration.

Q1: Ryan, congratulations on HELIX’s $800M valuation. What makes HELIX so different from other AI-powered trading strategies on the market today?

Ryan Pannell:

Thanks, Alicia; we’re pretty stoked with what the valuation firm came back with on HELIX. To answer your question, there are a few areas where HELIX stands out. For starters, it examines densely detailed volume compression at price patterns that immediately precede dynamic upside movement, and we don’t see anyone else doing that right now. 

To use an analogy that’s maybe more accessible, picture a buildup of pressure—water, gas, whatever you want to picture—inside a pipe that’s caused by a blockage. Initially, the blockage will cause whatever is flowing to slow or stop around the blockage, but ultimately the pressure increases to the point where the blockage is ejected—or the pipe bursts. With volume in a stock, growing density in both lot sizes and print execution in time series creates pressure which ultimately overcomes any selling that’s limiting upward movement at a specific price—and the result can be explosive. HELIX finds that pressure before the explosion, buys the stock, and then holds it just for a single day—just for the explosion—and then sells the position out the following close. And that’s the second novel part of the ideology: it’s very short-term. Not intraday, but close-to-close, so you have baked-in risk mitigation due to the trade duration. Once that alpha has been grabbed, HELIX is gone, and it doesn’t matter if there’s a subsequent pullback.

Q2: You’ve described HELIX as a “Synthetic Portfolio Manager” that’s fully AI-operated from end to end. Can you walk us through how HELIX works—from opportunity detection to trade execution?

Ryan Pannell:

Sure. Expanding on my previous answer, HELIX is looking for these volume compression patterns stuck at certain price points. Once it finds those, it considers the regime classification of the stock, the industry, the sector, the broad market, and compares those to what it’s seen before across millions of other real-world examples. We’re talking billions of concurrent examinations. 

Once it’s decided the candidate meets its ever-reevaluated weighted “buy” criteria, it takes the position. Once it has that, no matter what, it’s going to sell it the next day. Always; HELIX never holds longer than a day. Why? Because while you do miss the positive P&L from some longer runs, over the long term that’s a losing ideology. By focusing on just the explosion, we maintain a very high win rate over time, and that stacks up.

Q3: You’ve mentioned that volume compression patterns at price are foundational to Kaiju’s strategy. Why is this approach so effective, and how has it shaped your philosophy since Kaiju’s early days?

Ryan Pannell:

Of every criteria collection I have ever evaluated (and volume pressure at price is a “collection” of criteria, not just a single group), volume pressure is the most reliable. Think about it this way: what you are seeing when you watch price is a symptom, not a cause. Price doesn’t move up or down on its own; it’s responding to something. But what it’s responding to is where the focus needs to be. 

If you have a cough, for example, the cough is a symptom of something. Dry throat? Mild cold? Allergies? Flu? Lung cancer? Could be any of those things, but the treatment (and severity of consequence) is enormously different depending on which it is. Same for price. Maybe it’s moving because it’s hugely illiquid and the buyer (or seller) doesn’t know how to control their entry. Or maybe it’s moving because the company has a game-changing new technology that will reshape our world forever. Volume pressure only comes from one thing: enormous and growing aligned sentiment amongst market participants. And it only results in one thing… explosive movement. It’s not just a cause, it’s the cause.

Q4: The valuation firm’s methodology emphasized HELIX’s performance versus traditional S&P 500 strategies. Can you tell us more about the model’s 11-year CAGR and how the ARC® (AI risk containment) system factors into performance?

Ryan Pannell:

As with everything we build, we robustly test. We use blind (“walk forward”) model validations across real and synthetic markets (those are “real” markets that have had their time windows reordered—so “What if COVID appeared in the middle of 2016?”, for example). It’s why when we launch a strategy live, we’re never surprised by the results; everything operates on-model, all the time. 

With HELIX, the ARC® works in more of a forecasting mode than it does in other strategies where the time horizon is indeterminate (like DIP or BEX®). Because HELIX exits everything on the following close, it doesn’t need the ARC® to monitor positions to determine exits, but it can use the forward-looking synthetic scenarios the ARC® considers to gain confidence in its candidate selections.

Q5: I noticed your deep appreciation for your team—especially Dr. Muguruza and Dr. Žurič—in scaling HELIX. What makes their work so instrumental to this strategy’s evolution, and how does Kaiju attract such high-level talent?

Ryan Pannell:

The team is everything at Kaiju. I might invent the initial strategies, but it’s our incredible team that brings them to life underpinned by such power, and we couldn’t do anything we do without them. 

With HELIX, Aitor (Dr. Muguruza) immediately knew that Žan (Dr. Žurič) was going to click with it. That’s a huge part of how we do what we do; we don’t just “tell” a team member what they’re going to spend the next year working on—we ask “Who wants this one?” Everyone here has drivers that motivate them: some team members love to discover new exploitable patterns, but not necessarily drill down and flesh them out. Some love the challenge of taking a new discovery all the way through as far as we can take it, in the end being responsible for something truly formidable. 

It’s this process that I think keeps bringing us top talent, through our formal partnerships with Imperial College, ETH in Zurich, and now Oxford. Working in a dynamic, open, collaborative environment where you really have agency in what you take on and bring to life is hugely rewarding for people who do the kind of work we do, and I think a lot of shops miss that. 

HELIX might have been my idea, but under Aitor’s leadership and guidance, it definitely became Žan’s “baby.” By the time he was done with it, it was about 150% more powerful than anything I’d ever come up with, and operated substantially differently. It’s an amazing process to watch, really.

Q6: Beyond HELIX, Kaiju’s IP is now valued at over $1.45B, which is amazing. What’s next for your IP fund, and how do you envision predictive AI shaping other industries you’re exploring?

Ryan Pannell:

I think we’re done with the publicly accessible side of what we do. All our funds are closed to further subscriptions at this point, and we’re monetizing our assets, the results of which will be the achievement of all of our investment goals in each fund, and the returns speak for themselves. 

But running funds with outside investors is a lot of work, and I’ve been doing this for a long time now. I think we’re all interested in building strategies for ourselves a bit more (worked well for Renaissance Technologies), and potentially some outsourced portfolio management for larger asset managers and family offices. Sometimes we talk about another ETF for retail investors and RIAs, as that was pretty rewarding, or bringing AI candidate screening and risk mitigation to the retail trader, but we’ll have to see. For now, we’re focused on delivering value to our investors and crossing the finish line across all funds, and we’ll see how we feel after that. One thing is for sure: we’re here to stay.

Q7: You’ve long advocated for responsible AI—particularly the need for common standards and guardrails. How do you see predictive and generative AI working together to create more ethical, effective systems in financial services and beyond?

Ryan Pannell:

That’s a tough question to answer. Gen AI is always going to struggle with data standardization, and that’s by design; the source of its power is that it can consider “anything and everything” it can lay its hands on, while in predictive AI we use closed data sets. Ethically, generative AI struggles from “source of truth” challenges, while predictive struggles with behavioral manipulation challenges. 

Put them together and you have a potential catastrophe… Imagine a predictive system determining from your own behavioral data what misinformation you’re most likely to believe, and then a generative system creating that misinformation just for you. It gets scary, fast. 

On the flip side of that, if we want to keep it positive here, imagine a predictive behavioral system being able to assist you with thousands of everyday tasks that it knows you need to complete, and using a generative component for anything that requires novel creation. The result would be vastly less stress, and a huge net increase in time you could spend on all sorts of meaningful experiences. 

As always with AI, it boils down to us: AI isn’t inherently good or evil; it does what we tell it to.

Q8: For readers just entering the world of AI-based investing or predictive modeling—what advice would you give them as they look to build or evaluate new systems?

Ryan Pannell:

First, I’d say be realistic. The data you need are enormously expensive, and the processing power required to work with those data is equally so. Not as expensive as it was just 5 years ago, but the costs are still high. Then you’ve got the team you need to hire, and those folks are not easy to find nor are they cheap. Put it together and it’s a very daunting on-ramp—and then you have to add the lag, which is that you’re 5–10 years behind others like us who have been doing this for a while. You’ll never “catch up,” because we keep moving away. 

The second thing I’d say is in contrast to the first: don’t be afraid to be bold. The daunting on-ramp above is limited to real-time capital markets data consumption at the tick level, and billions of concurrent examinations, in support of micrometric trading decisions across the entire market. But you don’t need to do all that. Historical data isn’t expensive. Delayed data isn’t expensive (and can be free in some cases). Finding a repeatable pattern in a specific sub-set of stocks you know and like is not impossible with a single high-end computer, and a little programming knowledge goes a long way. 

It’s not unrealistic to think that you can, for little money, build a fairly powerful equities screening system that does a lot of heavy lifting for you, and performs examinations a human just simply cannot, in support of your own self-directed trading efforts. So if you’re considering getting started, just start. AI isn’t going anywhere, neither are capital markets, and we’re a long way from a “machine versus machine” trading future where you don’t stand a chance.

Ryan, thank you for offering this in-depth look at HELIX and Kaiju’s growing influence in the world of predictive AI. It’s always a pleasure to see how your team is combining innovation and responsibility—not just to drive returns, but to help shape a more intelligent and forward-thinking future for AI-driven investing. We look forward to what comes next.

Editor’s Note: This written interview was conducted by Alicia Shapiro, CMO of AiNews.com. The responses have been lightly edited for clarity and readability, while preserving the speaker’s original intent and voice. Structural formatting and editorial polish were supported by ChatGPT, an AI assistant. All final editorial decisions were made by Alicia Shapiro.