Precise and annotated 3D point cloud datasets are essential for training machine learning models in scene comprehension and indoor mapping. Nonetheless, obtaining high-quality real-world data is frequently resource-intensive and limited by equipment expenses, data imbalance, and the intricacies of annotation. This work presents a Python-based pipeline that systematically produces synthetic 3D point cloud data utilising geometric primitives—planes, spheres, and cubes—alongside adjustable noise, density, and semantic labelling. The platform facilitates the swift generation of labelled indoor scenes in PLY format, hence enabling data augmentation, benchmarking, and pre-training for deep learning models. Experimental findings illustrate the efficacy of the produced datasets in enhancing the generalisation of segmentation networks, corroborating algorithms with ground truth, and accelerating dataset curation in resource-constrained environments.