Accurately integrating manufacturing process outputs, such as void content, fiber volume fraction, and deformed shapes, into mechanical simulations is crucial for predicting the performance of composite structures. This work presents a robust multi-scale workflow that bridges the gap from micro-level properties to coupon-level and component-level mechanical analysis, while incorporating uncertainty quantification to improve predictive accuracy. At the micro level, a stochastic micro-mechanical model [1] (RVEtool) simulates Representative Volume Elements (RVEs) using high-performance computing (HPC) to capture manufacturing-induced variability, such as voids and fiber misalignment. The micro-level outputs are homogenized into effective material properties using a homogenization converter tool. This tool incorporates two complementary approaches: a Gaussian Process Regression (GPR) surrogate model for properties within defined ranges and a secondary converter for enhanced accuracy in outlier regions. At the coupon level, the homogenized properties are validated against experimental data and used in structural simulations to assess local material behavior under various loading conditions. Bayesian GPR models integrate uncertainties during training, generating predictive distributions that are more robust than deterministic surrogate approaches, which apply uncertainty quantification only after training. Finally, at the component level, the workflow translates manufacturing variability and uncertainties into mechanical simulations of full-scale composite structures. By propagating uncertainties from the micro to the component scale, this workflow rigorously accounts for defects, variations, and their cumulative effects on structural performance. Verified within the CAELESTIS project, this scalable, high-fidelity solution sets a benchmark for manufacturing-informed simulations, ensuring accurate performance predictions for composite components under real-world manufacturing-induced variability.