From Model to System

Initial Model (2023–2024)

• Strength: Good raw accuracy (~50% perfect / 69% directional).

• Weakness: Catastrophic -50% drawdown in Q3 2024, proving the need for disciplined risk management.

• Limitation: Token list included dead/delisted tokens, creating model drift risk.

Inteli-Trade Development (2024–2025)

• Training data: Nov 2023 – Aug 2024.

• Validation data: Aug 2024 – Aug 2025 (out-of-sample).

Optimization:

• Hyperparameter tuning via 13 parallel instances on GCP.

• K-fold cross-validation to prevent overfitting.

• Multi-metric scoring (profit, profit ratio, Sortino).

Risk controls implemented:

• Volatility-adjusted sizing.

• SMAPE prediction filters.

• Multi-exit strategies.

• Strict diversification + exposure limits.