Preliminary Round Participant Insights — AOT Matrix: Left-Brain Analysis, Right-Brain Decisions in AI Trading

Opening

In the WEEX AI Trading Hackathon, AOT Matrix chose a more cautious path in system design — one that’s actually harder to pull off in a live trading environment.

From the very start, they made clear choices about what role AI should and shouldn’t play in the trading system.

We interviewed AOT Matrix about their decision-making logic, the multiple iterations of their system architecture, and what it’s like to implement it under WEEX’s real trading environment and engineering constraints.

Q1. In AI trading, most people’s first instinct is “let AI place orders.” Why did you dismiss this idea from the start?

AOT Matrix:

Because crypto markets are inherently unstable.

Price distributions shift, volatility structures break, and historical patterns often fail when it matters most. Letting AI execute buy or sell orders directly would turn any model mismatch into immediate real losses.

Based on that, in the very first week we ruled out two common approaches: using AI as an automated trading bot, or letting it generate trading signals directly.

Instead, we chose to have AI answer a more restrained but far more critical question: is this the right environment to trade right now?

Q2. During the preparation phase, what system architecture did you initially experiment with?

AOT Matrix:

At first, we tried a hybrid setup: AI signals direction, and the rule-based system executes.

But during backtests and simulations, issues became clear: the stability of AI signals varied greatly across different market phases.

As soon as market structure shifted, the reliability of those signals dropped significantly.

We later realized the problem wasn’t the model accuracy — it was the very division of responsibilities.

Q3. How did you redefine the roles of AI and trading decision-making?

AOT Matrix:

After several iterations, we finalized a “left brain / right brain” system structure.

AI resides in the “left brain,” responsible solely for analysis and not for making trading decisions.

Its job is to assess market conditions — trending, ranging, high-risk scenarios, or whether trading should be paused — while providing a confidence score for the environment. It doesn’t predict exact prices or place orders.

Actual trading decisions are handled by the “right brain,” a rule-based system managing trade permissions, position sizing, and leverage controls.

Every trade must be auditable and replayable — a hard requirement we set for ourselves at the WEEX AI Hackathon.

Q4. During preparation, how challenging was it to translate trading experience into AI-readable input?

AOT Matrix:

Extremely challenging. Traders’ experience is often intuitive, but AI requires structured information.

So instead of just adding more data, we broke the logic down. We split trading logic into three types: market structure, volatility state, and risk conditions. AI learns and outputs only these intermediate states.

This way, AI no longer predicts future prices; it focuses on answering whether the current environment is healthy and suitable for trading.

Given the short preparation time, we believed this was a safer and more practical approach.

Q5. When integrating the WEEX API and moving from simulation to live trading, what unexpected challenges came up?

AOT Matrix:

Most challenges were engineering-related. We initially completed basic authentication and order submission via the WEEX API, but in live trading, we quickly realized that “being able to place orders” doesn’t guarantee long-term system stability.

Network jitter, request timeouts, and multi-strategy execution issues surfaced gradually during both simulations and live tests.

To fix this, we made systematic engineering upgrades, including:

• Full-chain trace IDs for order-level tracking
• Idempotent order controls to prevent duplicate executions
• Asynchronous queues and order status reconciliation to enhance system recovery under anomalies

This phase was a critical step in turning a demo into a system capable of long-term operation.

Q6. You put a lot of effort into recording trading decisions and executions. What was the reasoning behind this?

AOT Matrix:

In live trading, any trade that can’t be explained will eventually become a source of risk.

Therefore, we require that every order can answer three questions: Why was it opened at that moment? What did the system judge the market environment to be? Would the same decision hold if conditions repeated?

The system fully records AI assessments of market conditions, the rationale behind decision execution, and the final trade outcome.

The goal isn’t to complicate things, but to ensure all trades are traceable, replayable, and reviewable — what we call “full-chain auditability.”

Q7. While preparing for the WEEX AI Trading Hackathon, what has been your biggest insight about AI trading?

AOT Matrix:

Three main insights.

First, AI in trading is not meant to replace humans, but to constrain them.

It’s better at curbing emotional decisions and spotting untradeable environments than chasing “bigger returns.”

Second, system stability often matters more than model precision.

A system that looks perfect in backtests but fails live simply turns its technical edge into risk exposure.

Third, interpretability is critical for long-term survival.

Only if every P&L can be understood and reviewed can the system be fixed after drawdowns, rather than being scrapped and rebuilt.

Closing

For AOT Matrix, the WEEX AI Trading Hackathon isn’t just a model competition — it’s a comprehensive test of system design, engineering, and risk awareness.

Their architecture is the product of continuous validation, adjustments, and convergence under WEEX’s live trading conditions and engineering constraints.

And this is exactly the process AI trading must go through to move from concept to a sustainable, long-term tool.