Speaker diarization–the process of partitioning audio streams into segments associated with individual speakers–is a critical task in speech analytics, transcription systems, and human-computer interaction. This paper presents a comprehensive and modular speaker diarization pipeline that integrates state-of-the-art components to ensure high accuracy and scalability across diverse acoustic environments. The proposed system leverages Pyannote.audio for precise speech activity detection and segmentation, ECAPA-TDNN for robust and discriminative speaker embedding extraction, and SA-EEND (Speaker-Attributed End-to-End Neural Diarization) for intelligent speaker assignment, even in scenarios with an unknown number of speakers. Our pipeline addresses limitations in conventional systems by employing a hybrid framework that combines modular interpretability with the flexibility of end-to-end learning. Each component is independently fine-tuned and integrated to form a cohesive diarization system. Pyannote.audio ensures temporal segmentation with high recall, ECAPA-TDNN embeddings provide speaker-specific features with robustness against overlapping speech, and SA-EEND enhances attribution accuracy through attractor-based modeling. The system is validated on publicly available datasets with diverse speaker distributions and overlapping conditions. Evaluation across standard diarization metrics such as DER, JER, and SCER demonstrates notable improvements over baseline methods. The modularity of the system also enables future expansion toward multi-modal diarization and domain-specific applications.