Deception detection is crucial in domains like national security, privacy, judiciary, and courtroom trials. Differentiating truth from lies is inherently challenging due to many complex, diversified behavioural, physiological and cognitive aspects. Traditional lie detector tests (polygraphs) have been widely used but remain controversial due to scientific, ethical, and practical concerns. With advancements in machine learning, deception detection can be automated. However, existing secondary datasets are limited-they are small, unimodal, and predominantly based on non-Indian populations. To address these gaps, we present CogniModal-D, a primary real-world multimodal dataset for deception detection, specifically targeting the Indian population. It spans seven modalities-electroencephalography (EEG), electrocardiography (ECG), electrooculography (EOG), eye-gaze, galvanic skin response (GSR), audio, and video-collected from over 100 subjects. The data was gathered through tasks focused on social relationships and controlled mock crime interrogations. Our multimodal AI-based score-level fusion approach integrates diverse verbal and nonverbal cues, significantly improving deception detection accuracy compared to unimodal methods. Performance improvements of up to 15% were observed in mock crime and best friend scenarios with multimodal fusion. Notably, behavioural modalities (audio, video, gaze, GSR) proved more robust than neurophysiological ones (EEG, ECG, EOG).The study demonstrates that multimodal features offer superior discriminatory power in deception detection. These insights highlight the pivotal role of integrating multiple modalities to develop robust, scalable, and advanced deception detection systems in the future.