The AI Engines

A deep reinforcement-learning agent (Deep Q-Network family). It was trained by reward and penalty over enormous numbers of simulated trading decisions, learning a policy that maps market state to the action — buy, sell, or hold — with the best expected outcome. In production it runs in inference mode: it applies what it learned, it is not gambling on live experiments.

A stacked ensemble of neural networks focused on short and mid-term price forecasting. It engineers dozens of features from recent price and volume action and produces a directional probability — its vote — together with a confidence that reflects how familiar the current setup looks.

A quantum-inspired probabilistic simulator. For each decision it runs thousands of Monte-Carlo simulations over possible near-future market paths, encoding market state into a statevector representation. Its vote reflects which outcomes dominate the simulated distribution — and how decisively.

A transformer — the same architectural family behind modern language models — applied to market time series. Attention layers let it weigh which past moments matter for the present, recognizing regime-dependent temporal patterns that fixed-window models miss.

A convolutional network that reads candlestick charts the way a trained chartist does — but across every asset, on every cycle, without fatigue. It detects accumulation, breakouts, exhaustion, and reversal structures and votes on the direction they imply.

The Coherence Model doesn't vote on direction. Using spectral analysis (frequency-domain methods), it measures how aligned price action is across assets and timeframes — whether the market is moving as a coherent field or as fragmented noise. Its output scales the ensemble's confidence: in a chaotic field, even a unanimous vote is treated more cautiously.