Alzheimer's disease is typically diagnosed using invasive and costly imaging and biomarker-based approaches, often after significant neurological decline has occurred. This work proposes an AI-driven multimodal framework that integrates speech features, neuropsychiatric symptoms measured using the Neuropsychiatric Inventory Questionnaire (NPI-Q), and cognitive assessments for Alzheimer's disease risk stratification. The framework consists of two algorithms: the first independently estimates stage probabilities using Random Forest classifiers, while the second combines modality-specific probabilities through a weighted probabilistic fusion scheme. This framework is designed to support classification of Cognitively Normal (CN), Mild Cognitive Impairment (MCI), and Alzheimer's Disease (AD), while accommodating incomplete and asynchronous multimodal data. Future work will validate the framework using real-world longitudinal datasets and optimize modality-specific weights to improve predictive performanc