Investment Strategy
Deep learning architecture purpose-built for XAUUSD. Trained on historical data, validated forward, executing live.
Model Architecture
The core engine is a Temporal Fusion Transformer (TFT) — a state-of-the-art deep learning architecture that integrates LSTM for sequential processing, multi-head self-attention for long-range dependencies, and quantile outputs for uncertainty estimation.
64
Hidden Size
3
LSTM Layers
4
Attention Heads
7
Quantile Outputs
Data Pipeline
Trained on pre-processed historical OHLCV data, split chronologically (80/20) with zero data leakage. Features computed in real-time during live execution mirror the exact training pipeline.
Technical Indicators
ATR for volatility, 12/26-period EMAs, MACD + signal line for momentum, RSI for overbought/oversold detection.
Custom RSI-KDE Signals
Gaussian Kernel Density Estimation on RSI pivots generates proprietary bullish/bearish probability densities.
Lagged Close Prices
5-step lagged close values capture short-term autocorrelation during trending and mean-reverting phases.
Chunk-Based Generalization
Data segmented by chunk_id prevents regime-specific memorization across different market conditions.
Decision Engine
Live trading loop fetches real-time M1/M5 bars and runs inference through the trained TFT. Median quantile drives direction, rate-of-change analysis adds momentum context.
Data Fetch
10,000 bars + live feed
Preprocess
Compute indicators
Inference
7 quantiles, 10 steps
Signal
Median + ROC analysis
Execute
Filtered order with SL/TP
Risk Management Architecture
Position Limits: Max 4 concurrent trades. Opposite-direction blocking prevents conflicting positions.
Stop-Loss / Take-Profit: Fixed 300 pip SL, 600 pip TP — consistent 2:1 reward-to-risk on every execution.
Trailing Stop: Activates at +400 pips, trailing 130 pips behind to lock gains while letting winners run.
Dynamic Lot Sizing: Base 0.1 lots per $1,000. Halves at -5% weekly, quarters at -15%. Conservative scaling on profits.
Built to Outperform.
Not to sell courses. Not to collect subscriptions. If we don't perform, we don't eat.
Apply for Founding Membership →