Global tariffs, ranging from 10% to over 100% in 2025, alongside geopolitical uncertainties from international elections, are expected to heighten economic volatility, increasing the risk of consumer loan delinquency for fintech lenders. This paper proposes an AI-driven credit risk assessment framework that integrates machine learning (XGBoost), deep learning (long short-term memory networks), and alternative data to dynamically predict repayment risks. Implemented using open-source tools (scikit-learn, TensorFlow, and XGBoost) in Google Colab, the framework achieves a 12% higher default prediction accuracy (LSTM ROCAUC: 0.82 vs. XGBoost: 0.73) and reduces potential losses by 10%. A reproducible Python script and process flow diagrams validate the approach, providing a robust, privacy-compliant solution for fintech lending resilience in tariff-impacted environments. The framework addresses data privacy and regulatory compliance, ensuring practical applicability.