We present a mechanistic multi-scale modeling framework that links within-host viral kinetics to population-level transmission in order to explain variant-specific epidemic dynamics of SARS-CoV-2. The model combines a within-host target-cell-limited viral dynamics system with a between-host agent-based transmission model on a structured contact network. Variant-specific viral load trajectories are inferred from longitudinal Ct data using hierarchical Bayesian inference and then translated into time-varying infectivity through a viral-load-based transmission mapping. This allows infectiousness to vary over time according to biologically grounded kinetics rather than fixed transmission parameters. To isolate the effect of temporal infectiousness, the multi-scale model is compared with a baseline model calibrated to the same reproduction number. Simulation results show that differences in viral proliferation and clearance substantially alter epidemic trajectories even under matched overall transmissibility. Faster viral expansion leads to earlier and sharper epidemic peaks, whereas slower growth and prolonged clearance produce delayed and broader outbreaks. The model also generates heterogeneous secondary-case distributions and occasional high-impact transmission events without imposing ad hoc superspreading assumptions. These findings highlight the importance of coupling within-host and between-host processes in epidemic modeling and suggest that viral kinetics play a central role in shaping epidemic timing, burden, and transmission heterogeneity.