TradeVision™/Neural Network
🧠
TradeVision™ AI Architecture

Neural Network Hub

CrossMamba™ • RangeVision™ • ForeSight™ MA • XGBoost v4

19:06:024 ENGINE TYPES • 72+ MODELS
Total Models
108
XGBoost
72
CrossMamba
36
CM Wins
0/36
Best Accuracy
0.0%
GPU Cost
~$0.20/wk
⚡

CrossMamba™

TRAINED

Hybrid Mamba-Transformer direction prediction

Models
36
Wins
0/36
Best
0.0%
Avg
0.0%
Stream 1: Mamba (720 candles, long-range). Stream 2: Transformer (120 candles, short-range). Fusion: Learned gating. Binary UP/DOWN + 0-100 strength.
🎯

RangeVision™

NEXT PHASE

PatchMamba predicted high/low price range + AI levels

Models
0
Type
Regression
Output
T1/T2/SL
Arch
PatchMamba
Bi-directional Mamba + Patch Attention. Predicts exact price levels (T1, T2, T3, SL) instead of rule-based structural zones. Trained on historical target hit/miss data.
🔮

ForeSight™ MA

NEXT PHASE

Temporal Fusion Transformer predictive moving averages

Models
0
Type
Multi-Horizon
Output
5/20/50 MA
Arch
TFT
Predicts where 5, 20, 50-period moving averages WILL BE 1-3 periods ahead. The blue line leads price instead of lagging it. Uses interpretable attention weights.
🌳

XGBoost v4

PRODUCTION

Gradient Boosted Trees with 194 cross-market features

Models
72
D1 Avg
56.5%
H1 Avg
54.4%
Features
194
88 single-market + 106 cross-market features. 16 statistical tests (Pearson, Granger, MI, HMM). Walk-forward validated. Weekly auto-retrain.
AI Architecture Flow
📊
Market Data
M15, H1, H4, D1 17 assets
→
⚙️
Feature Engine
194 features 88 market + 106 cross
→
🧠
AI Models
XGBoost + CrossMamba Best-per-asset selection
→
🏗
Structural Engine
Zone detection Dynamic ATR minimums
→
🎯
3-Card Output
SHORT / DAY / SWING T1, T2, T3 + SL
GPU Training: RunPod A40 ($0.39/hr) • Weekly auto-retrain Saturday 4AM UTC • Models versioned and A/B tested • Only deployed if accuracy improves