SignalGPT: A Deployable and Interpretable AI Framework for 24-Hour Directional Market Signal Prediction

Short-horizon financial market prediction remains a challenging problem due to the inherent noise, non-stationarity, and partial efficiency of price movements. While numerous studies have explored long-term forecasting and high-frequency trading strategies, comparatively limited attention has been given to deployable, interpretable systems for short-term directional prediction that are accessible to independent researchers and practitioners. This paper presents SignalGPT, an end-to-end artificial intelligence framework designed to predict 24-hour directional market signals using a ternary classification scheme comprising UP, DOWN, and NEUTRAL outcomes. The system integrates trend-aware feature engineering with supervised machine learning models to produce probability-based predictions, explicitly exposing model confidence rather than deterministic forecasts. A neutral class is introduced to account for uncertainty and low-confidence regimes, reinforcing responsible probabilistic reasoning. SignalGPT is implemented as a modular, deployable architecture consisting of a data ingestion layer, feature extraction pipeline, model inference service, and a lightweight web-based conversational interface. Publicly available financial time-series data are used to ensure reproducibility, and the complete system is released as an open-source research prototype with a publicly accessible deployment. The proposed framework demonstrates the feasibility of building transparent, low-latency predictive systems for short-horizon financial signals without reliance on proprietary data or infrastructure. This work contributes a reproducible, interpretable AI research artefact suitable for further academic investigation and real-world demonstration.