Dementia affects over 55 million people globally and costs the United States alone an estimated $345 billion annually. Yet the most widely used diagnostic tools-the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA)-were designed decades ago, require trained clinicians to administer, and focus primarily on memory recall. They are blind to one of the earliest and most information-rich signals of cognitive decline: language. This paper presents COMMA (Cognitive Oral Marker and Monitoring Assessment), an AI-powered speech assessment platform developed to address this gap. COMMA analyzes everyday spontaneous speech across three linguistic dimensionscorrectness, fluency, and syntactic complexity-using a pipeline built on OpenAI Whisper, spaCy, NLTK, and PyDub. We describe the theoretical basis for speech-based cognitive assessment, the technical architecture of the COMMA scoring system, its differentiation from existing tools, and its potential to transform early dementia detection in primary care and home settings across the United States. The proposed approach demonstrates how linguistic biomarkers can enable scalable, low-cost cognitive screening that complements traditional neuropsychological assessment.