From Black Box to Glass Box: Deep3’s Personalized AI Trading Platform

Making AI Trustworthy in Financial Decision-Making

Deep3’s mission is to make AI and machine learning accessible as core components of the web3 user experience by making it easy for developers to deploy and simple for users to control. As a data science team with decades of combined experience, they’ve identified a critical missing component in the Web3 ecosystem experience: personalization.

“Personalization is what created Web2 as we know it today—it’s what brought your grandma online,” explains Daniel Stephens, founder of Deep3. “Web3 hasn’t figured this out yet. Every time you connect to an app or do anything on a Web3 platform, you’re generally getting the exact same experience as everyone else.”

Deep3’s flagship application, Hōkū, aims to change this by analyzing each wallet’s trading history with advanced AI models and monitoring network activity in real-time to instantly deliver personalized token and dapp recommendations, while their visual 3D network explorer reveals patterns that even the most advanced block explorers can’t show. 

However, as the platform evolves from making recommendations to automating trades, a fundamental challenge emerges:

How do you build trust when AI is making financial decisions with users’ assets?

This challenge is magnified when the AI is not only analyzing data but directly controlling users’ wallets:

• Access Key Risks: Granting an AI access to private keys or transaction signing introduces new attack vectors and requires bullet‑proof key custody solutions.
• Regulatory and Compliance Complexity: Ensuring AI‑driven transactions comply with KYC/AML rules and evolving crypto regulations demands continuous monitoring and auditability.
• Explainability of Actions: Users need clear, human‑readable rationales for why the AI bought, sold, or held an asset—especially when real money is on the line.

“When we move into the realm of not only our code managing your money, but an AI that we’ve built managing your money on top of that, there’s so many elements of that stack that need an extra level of transparency and verifiability to gain user confidence,” explains Stephens.

Verifiable Guardrails for AI Decision-Making

After evaluating verification solutions, Deep3 chose to integrate Truebit Verify’s transparent verified compute platform to build trust in Hōkū’s AI-driven trading functionality.

What particularly attracted Deep3 to Truebit Verify was Truebit’s approach to decentralized verification:

“The approach of decentralizing the decision consensus process is really attractive for us because we think it ultimately is the right source to scale verifiability. If you wanted to sit down and explain to a layperson how you ensure that this decision the software just made is the right one, the way that Truebit Verify is built on a fundamental level is just easier to understand and explain. That’s really attractive when AI is inherently opaque—any opportunity we have to do things that are more transparent, more understandable to people gets a lot of consideration from us.”

This integration manifests in several key ways:

1. Verifiable Parameter Enforcement

As Hōkū evolves to become an autonomous trading agent, users will set key guardrails like maximum position sizes and other risk management parameters. Truebit Verify will provide cryptographic proof that the AI respects these parameters.

For example, if a user specifies that no position should exceed 20% of their portfolio value, Truebit Verify will create a transparent, verifiable record proving the AI adheres to this constraint without requiring users to manually verify calculations.

2. Creating Trust in Complex AI Decisions

When Hōkū’s AI identifies potentially profitable trades (which it does daily), users need confidence that the underlying analysis is sound. By leveraging Truebit’s verification layer, Deep3 can provide assurances about the integrity of these recommendations without compromising the sophisticated algorithms that power them.

3. Building for Future Verification Needs

Beyond immediate trading applications, Deep3 is laying groundwork for more comprehensive AI verification:

“Long term, I absolutely think you’re going to find a very large market for truly verifiable AI algorithms where the core of the computation process—whether it’s inference or training—needs to have that verifiability,” notes Stephens. “An AI that’s operating inside of a pacemaker isn’t science fiction. It’s going to happen. But you better believe that people are going to want a higher degree of transparency and verifiability when someone’s heart is being controlled by an AI.”