Here we look to create a holistic big-data forecasting pipeline in the financial domain, which connects the macroeconomic indicators with market level decision systems. Relying on deep learning models and data preprocessing techniques, we present a scalable model framework to support variety of diverse datasets including but not limited to macroeconomic indicators, financial time series and sentiment data for forecasting financial market movements. The pipeline consists of five main modules, data collection, feature cleaning, cavity feature extraction model training and auto-evaluation. Empirical results show that the proposed approach can lead to remarkable improvements on the accuracy, robustness and timeliness of prediction than the traditional econometric method. This study does not only contribute to the methodological investigation of data-driven financial forecasting, but also offers technical wisdoms for developing intelligent financial analytics platforms in the new era of digital finance.