One of the challenges in building reaction mechanisms using algorithms for automated model construction is to describe the essential chemistry and enable prediction of experimental data over wide ranges of reaction conditions while maintaining a manageable model size. Two complementary methodologies for building compact reaction mechanisms were developed and combined with existing algorithms based on graph theory and bond-electron matrix operations. Each strategy was developed using pentadecylbenzene pyrolysis as an illustrative example. The first approach used a radical lumping strategy to group radicals according to their reactivity. The mechanism was reduced from 719 to 215 species and successfully predicted the experimentally observed initial reactivity. However, implementation of the radical lumping criteria alone was insufficient to allow for secondary reactions to higher-rank products. Therefore, on-the-fly sensitivity analysis was incorporated to identify the important and necessary species as the mechanism was generated to guide the mechanism building process. The generic algorithms developed can be applied to generate compact reaction mechanisms for a wide away of higher molecular-weight reactants.