Forecasting stock price movements is a complex task due to inherent market volatility, nonlinear dependencies, and inter-stock correlations. This paper proposes a cluster-informed deep learning framework for the simultaneous prediction of the next-day opening price direction across 46 constituents of the NIFTY 50 index. Unlike traditional single-stock models, our approach leverages shared sectoral patterns by integrating unsupervised similarity-based clustering with a unified multi-output architecture. We first identify co-movement relationships among stocks using a combination of correlation and cosine similarity metrics, followed by hierarchical clustering. These relationships inform the design of a hybrid neural model: a Time-Distributed dense encoder for dimensionality reduction, parallel LSTM layers for temporal representation learning, and a shared LSTM block leading to stock-specific output heads. Extensive experiments demonstrate that the proposed model consistently outperforms statistical and rule-based baselines in directional accuracy and the F1-score. The results highlight the efficacy of incorporating sector-level dependencies and structured temporal learning in robust financial forecasting.