Researchers constantly strive to explore large and complex search spaces in various scientific studies and physical experiments. These could include searching the exponentially large set of amino-acid combinations to form a drug that could potentially serve as a medicine for a target disease. Experiments could range from transistor characterization at the nanoscale to astronomical measurements. Such investigations often involve sophisticated simulators or time-consuming experiments that make exploring and observing new design samples challenging. To address these limitations, we propose a differentiable surrogate-based modeling and optimization approach that minimizes the number of samples that need to be explored.