This study introduces a novel multi-modal deep learning framework for predicting consumer credit scores by integrating structured financial data from credit reports with unstructured textual narratives. The proposed architecture employs two parallel convolutional submodels that independently process numerical and textual inputs. These outputs are then merged to capture cross-modal feature interactions. Experimental results show that the proposed model significantly outperforms traditional machine learning approaches (XGBoost, Random Forest) based on metrics including RMSE, R \(^{2}\) , F1 score, and AUPRC. The work underscores the value of using qualitative textual data alongside quantitative data to enhance the assessment of creditworthiness, particularly in cases with limited financial history. This approach contributes toward more inclusive and accurate credit assessment systems for emerging economies.