Oncogenic RAS mutations, which are common in human tumors and occur in about 30 % of cancer cases, present significant challenges to effective cancer treatment. Among the KRAS family, the KRAS-G12D mutation is a promising target for treating different types of cancer. Current approaches to inhibit the KRAS-G12D mutation have shown limited success, highlighting the urgent need for innovative therapies. In this study, we employed machine learning, followed by scaffold and core hopping fragmentation, to design, synthesize, and biologically test several 1-oxa-3,7-diazaspirodecane-2-one compounds, ultimately identifying two new KRAS-G12D inhibitors. Multiple in silico evaluations were performed to explore the potential of these inhibitors and to gain a deeper structural understanding of how these compounds bind within the KRAS-G12D active site. Additionally, protein binding assays and other biological tests demonstrated that these compounds exhibit a strong protein binding affinity (Kd of 28.29, 48.17, and 85.17 nM) and high selectivity for KRAS-G12D. Subsequent cellular assays further prioritized HDB-2 and HDB-3 as potent KRAS-G12D inhibitors, each showing nanomolar IC50 values. These results suggest that these compounds could become highly effective and selective anticancer agents for targeting KRAS-G12D-driven tumors.