In the crowded cellular environment, the transcriptome is highly organised with a network of interactions between proteins and RNA molecules that are crucial for the formation of biomolecular condensates. While the role of proteins in this process has been extensively studied, the contribution of RNA has remained underexplored. In this study, we identify a distinct set of RNAs that are prone to condensation during early embryonic development and apply an integrative deep learning approach to deconvolve the sequence and structural determinants, and trans-acting regulators involved in RNA condensation. These RNAs form a highly interconnected subnetwork of RNA-RNA interactions with distinct local connectivity patterns, and encode proteins with intrinsically-disordered regions and a propensity for phase separation. We find that although the identity of condensation-prone RNAs differs across developmental stages, their organisational principle remains conserved. This suggests a multi-layered coordination of genes driving both RNA and protein phase-separating processes. This work highlights the unique organisation of transcripts encoding proteins with similar biophysical properties and provides insights into the principles governing RNA assembly within biomolecular condensates during early embryonic development.