Transition states (TS) are important molecular conformations with applications in drug development, materials science and biology. The transient nature of these conformations makes studying them extremely challenging. Simulating these states requires expensive computational algorithms whose convergence is never guaranteed. Machine learning (ML) has led to significant improvements in various domains, including computational chemistry. In this thesis, we propose a novel ML pipeline for sampling TS in molecular systems. We combine the expressive capability of graph neural networks to encode graph representations of molecular conformations which is then used for training, with generative models that are able to predict TS structure as output. Using our proposed method, we successfully demonstrate the sampling of minimum energy pathways and conformational TS in alanine dipeptide in one-shot without the need for any example TS or reaction coordinates. Our unsupervised novel ML approach shows promise in TS search and could lead the path for further innovations and analysis of more complex chemical systems.