The rapid proliferation of AI technologies has significantly impacted diverse fields, particularly finance. However, many AI models face challenges due to their high computational demands and impracticality for real-time deployment. In this study, we propose a lightweight framework to integrate Bidirectional Long Short-Term Memory (BiLSTM) networks into existing automated forex trading systems, enabling real-time filtering of suboptimal trades. Empirical results demonstrate that the proposed system achieves an 70% reduction in ineffective trades compared to the baseline framework.