This paper presents APEX AI, a professional-grade equity trading signal platform designed for National Stock Exchange (NSE) listed Indian stocks. The system employs a heterogeneous ensemble of three complementary machine learning models: Gated Recurrent Unit (GRU) networks for sequential pattern capture, Temporal Convolutional Networks (TCN) for multi-scale temporal feature extraction, and LightGBM for gradient-boosted tabular learning. These models are fused through a soft-voting ensemble to produce probabilistic price forecasts expressed as P10, P50, and P90 quantile estimates over a 14-day horizon. A four-stage gate architecture governs signal quality, filtering signals based on trend alignment, volatility regime, volume confirmation, and risk-adjusted expected return. The platform exposes predictions through a FastAPI backend and a React/TypeScript/Vite frontend featuring a TradingView-style candlestick chart with an integrated forecast cone. Experimental evaluation on historical NSE data demonstrates directional accuracy above 62%, with the ensemble outperforming any individual constituent model.
Ghodke et al. (2026) studied this question.