Additive manufacturing (AM) has shown great potentials in fabricating titanium aluminides (TiAl-based alloys) toward high-temperature components in aerospace and automotive applications. However, due to the complex thermal conditions during AM, the as-printed components typically contain heterogeneous microstructure, leading to nonuniform mechanical properties. A thorough understanding of microstructure evolution during AM is necessary to fabricate high-performance TiAl-based components. In this work, the mechanism for the formation of heterogeneous microstructure during selective laser melting (SLM), particularly the spatial variations in sub-grain cellular structure, was revealed by a computational framework. Specifically, a binary Ti-45Al (at.%) alloy was used for the SLM experimental observation and model development to investigate the process-microstructure relationship. The computational framework integrates a finite element thermal model and a phase-field microstructural model. A particular focus was put on the local sub-grain cellular structure evolution within the melt pool. The microstructural sensitivity to spatial variations and individual processing parameters were investigated to better understand the non-equilibrium solidification during SLM. Good agreements in the sub-grain size were achieved between experimental measurements and modeling predictions. This work presents valuable insights and guidance toward the process optimization and alloy design for fabricating high-performance TiAl-based alloys.