Riverbank erosion remains a critical challenge in civil engineering, characterized by highly nonlinear interactions between hydraulic forces, soil properties, and channel geometry. Traditional linear regression approaches often fail to capture these complex dynamics, particularly under varying flow conditions. This keynote presents an interdisciplinary adventure—transplanting the Neural Network AutoRegressive eXogenous (NNARX) model, a technique rooted in control systems engineering, into the domain of geotechnical and hydraulic prediction.

Using 338 field measurements from Sungai Bernam, Malaysia, collected over 12 months across both low and high flow conditions, we demonstrate how NNARX successfully predicts riverbank erosion rates with R² values of 0.932 (training) and 0.788 (testing), achieving 94% accuracy within acceptable discrepancy limits. Sensitivity analysis identified the ratio of shear velocity to near-bank velocity as the most influential parameter, revealing counter-intuitive findings: maximum erosion (1.5–1.8 m/year) occurred at lower velocities (0.2–0.5 m/s), challenging conventional assumptions about flow-erosion relationships.

Beyond model development, this talk showcases a second AI frontier: leveraging large language models (LLMs) and prompt engineering to accelerate research workflows. We demonstrate how AI tools transformed raw survey data from 1,029 respondents into publication-ready statistical analysis, generating reproducible Python scripts, performing robustness checks, and drafting structured academic narratives—compressing weeks of analytical work into days. This keynote offers civil engineering researchers’ practical insights into two converging AI applications: adopting cross-disciplinary machine learning techniques for complex physical predictions and harnessing generative AI as a research accelerator. Both approaches represent accessible entry points for practitioners seeking to integrate AI into their research toolkit without requiring deep expertise in computer science.