Polymer nanocomposites (PNCs) combine polymer matrices with nanofillers to achieve exceptional mechanical, thermal, and functional properties. However, understanding how inter-nanoparticle (NP) repulsive interactions influence structural assembly and transport properties in confined domains remains under-researched. In this study, we employ GPU-accelerated Langevin dynamics simulations to systematically investigate how varying inter-NP repulsive strength drives phase behaviour in strongly confined PNC systems. Semi-flexible polymer chains and spherical NPs were simulated in extremely confined (quasi-2D) geometries with fixed monomer and NP volume fractions. Structural properties were studied via radial distribution function, cluster analyses, crystallinity and nematic order parameter calculations, while diffusivity was assessed through mean-square displacement analysis. Depending on the strength of repulsion, the composite system self-assembles into distinct morphologies: i) a droplet phase, characterised by complete NP phase separation; ii) a slab phase featuring percolating networks and a maximum nematic order; iii) a broken-slab phase showing polymer infiltration and fragmented assemblies; and iv) a dispersed phase with ordered NP distribution, exhibiting increased smecticity at higher repulsion strengths and crystalline spatial organisation as confirmed by RDF and crystallographic analysis. Diffusivity exhibits non-monotonic behaviour, with optimal nanoparticle mobility at intermediate repulsive strengths, while polymer diffusivities show inverse trends indicating competing dynamics between the two components. These findings demonstrate that fine-tuning inter-nanoparticle repulsions provides precise control over structural phases and transport properties, offering design principles for PNCs with the required functionality.