The development of deep learning based image segmentation algorithms is often hindered by the lack of adequately annotated datasets, and this issue becomes a severe bottleneck in multi-class segmentation. The models can learn binary image segmentation since the datasets are often labeled for a single class. However, the task becomes more challenging when it comes to multi-class annotations and segmentation of regions of interest, e.g. in multi-organ segmentation that plays a vital role in computer assisted diagnosis, surgery planning, and other related applications. This study proposes a framework, CabiNet, that attempts to solve data bottleneck by learning multiclass healthy organ segmentation from low amount of partially annotated abdominal MRI data. Multiple organ-specific expert networks and a Jack of All (JoA) network are learned to generate posterior probability maps of individual organs simultaneously which are accumulated to get final posterior probability distributions over different organs for each pixel. On benchmark Combined Healthy Abdominal Organ Segmentation (CHAOS) MRI dataset CabiNet yielded very promising dice scores: liver (90.39%), right kidney (87.41%), left kidney (81.09%), spleen (90.78%) and overall average dice score as 87.4%.