Abstract
The rapid growth of chemical space and the high cost, long timelines, and low success rates of traditional drug discovery pipelines present significant challenges to pharmaceutical research. Conventional approaches rely heavily on trial-and-error experimentation, extensive laboratory synthesis, and time-consuming screening procedures, which collectively slow down the development of effective therapeutics. In response to these limitations, this paper presents Generative Intelligence for Synthetic Drug Discovery and Molecular Generation, an AI-driven framework that integrates deep generative modeling, molecular property evaluation, and AI-assisted synthesis planning to accelerate early-stage drug discovery. The proposed system leverages advanced machine learning techniques, particularly Large Language Model(LLM), to learn meaningful latent representations of molecular structures from large-scale chemical datasets. By encoding and decoding molecular information using SMILES representations, the model is capable of generating novel, chemically valid, and diverse molecular candidates that expand beyond known chemical libraries. The generative intelligence framework ensures that the produced molecules adhere to fundamental chemical constraints while maintaining structural novelty and diversity, thereby addressing the exploration-exploitation trade-off in molecular design. To enhance practical applicability, the system incorporates automated molecular property prediction and drug-likeness evaluation using established pharmacokinetic and physicochemical rules, including Lipinski's Rule of Five and ADMET-related constraints. These evaluations act as intelligent filters that discard infeasible or unsafe compounds at an early stage, reducing downstream experimental costs.