This project develops generative diffusion and flow matching frameworks that operate natively in discrete state spaces via continuous-time Markov chains. By bypassing continuous relaxations, the research addresses the algorithmic challenges of efficient sampling and transition dynamics. Crucially,