<p>Accurate assessment of part complexity in Additive Manufacturing (AM) is critical for design for AM, cost estimation, and risk mitigation; however, the Analytic Hierarchy Process (AHP) methods currently in use are slow, subjective, and not realistic at an industrial scale. We present an AHP-informed AI framework that can learn to replicate expert AHP scoring using supervised machine learning. A dataset of 100 parts from two different real-world systems was compiled from expert-scored AHP notebooks and then trained on multi-modal descriptions of each part to predict complexity. The Random Forest Regressor model successfully established good predictive fidelity on hold-out parts (R² = 0.897; MAE = 8.533; RMSE = 10.519, expressed in native AHP-score units), that was a significant improvement on baseline regressors. Feature-importance analysis indicated agreement with hierarchy criteria maintained by human experts, as well as robustness to empty values and unseen mode categories. A deployable desktop application operationalizing the pipeline was developed that enables engineers to obtain predictions in milliseconds or lower timeframes and reduces the time to evaluate a part by more than 99% compared to manually scoring the part. There remains a gap connecting structured expert judgment with evidence-based inference; we have piloted a scalable, interpretable, and deployable technology for specifying part complexity in AM workflows.</p>