AI Learning Data System: A Structure That Automatically Collects All Trade Signals
December 2025
Purpose of the learning data system
NoahAI Labs’ financial AI aims for not “fixed automation” but “a judgment structure that accumulates experience.” To that end, we built a learning data system that automatically collects and analyzes every trade signal.
The goals of this system are as follows.
- Continuous learning: Ongoing learning grounded in real trading experience
- Pattern recognition: Automatic recognition and learning of market patterns
- Dynamic optimization: Dynamic tuning as markets change
- Per-exchange characteristics: Independent learning for each exchange’s behavior
Data collection architecture
1. Automatic signal collection
We collect signals automatically on every exchange on a one-minute cadence.
- APISignalManager: Per-exchange API signal exchange and learning data generation
- Automatic execution: Runs in the background without manual steps
- Normalization: Converts each exchange’s data into a standard format
- Persistence: Automatic storage in the learning database
2. Types of collected data
- Market data: Price, volume, volatility, and more
- Technical indicators: RSI, MACD, Bollinger Bands, etc.
- Market sentiment: Fear & greed indices, funding, etc.
- Trade signals: Buy/sell signals and outcomes
Design of the data structures
1. Signal history
We record the history of every trade signal.
- Signal timestamp
- Signal type (buy / sell)
- Market regime (up / down / sideways)
- Trade result (P&L)
2. Pattern analysis
We analyze collected data and extract patterns.
- Successful trade patterns
- Unsuccessful trade patterns
- Patterns by market condition
- Patterns by exchange
3. Dynamic thresholds
We adjust thresholds dynamically as markets change.
- Thresholds by market regime
- Thresholds by exchange
- Thresholds by time window
- Automatic adjustment mechanisms
4. Exchange-specific behavior
We learn each exchange’s characteristics independently.
- Per-exchange patterns
- Per-exchange optimization strategies
- Per-exchange risk profiles
Learning loop structure
1. Record
Every trade signal and outcome is logged in a standard format.
2. Normalize
Per-exchange data is converted into a standard format.
3. Insight
Logged data is analyzed to extract patterns and insights.
4. Policy
Based on extracted insights, we update trading policy.
5. Risk
Updated policy is validated from a risk perspective.
6. Feedback
Actual trade results are fed back into the next learning cycle.
Key characteristics
1. Automation
Every step runs automatically; no manual intervention is required.
2. Persistence
Learning continues for as long as the system runs.
3. Independence
Each exchange is learned independently, reflecting its own behavior.
4. Verifiability
All learning data is stored in a standard format and can be audited.
Government R&D and investor perspective
Technical innovation
The AI learning data system enables continuous learning and automatic optimization.
- Automatic learning: Learns without manual intervention
- Pattern recognition: Automatically recognizes market patterns
- Dynamic optimization: Tunes behavior as markets change
Extensibility
The learning data system continues to work as new exchanges or asset classes are added.
- When a new exchange is added, learning starts automatically
- When a new asset class is added, it learns independently
- Existing learning data is not unnecessarily disrupted
Conclusion
The AI learning data system is a core system that realizes not “fixed automation” but “a judgment structure that accumulates experience.”
Automatically collecting and analyzing every trade signal, then learning and optimizing continuously, is an important example of technical innovation and extensibility.