We address the problem of designing micro-fluidic chips for sample preparation, which is a crucial step in many experimental processes in chemical and biological sciences. One of the objectives of sample preparation is to dilute the sample fluid, called reactant, using another fluid called buffer, to produce desired volumes of fluid with prespecified reactant concentrations. In the model we adopt, these fluids are manipulated in discrete volumes called droplets. The dilution process is represented by a mixing graph whose nodes represent 1-1 micro-mixers and edges represent channels for transporting fluids. In this work we focus on designing such mixing graphs when the given sample (also referred to as the target) consists of a single-droplet, and the objective is to minimize total fluid waste. Our main contribution is an efficient algorithm called RPRIS that guarantees a better provable worst-case bound on waste and significantly outperforms state-of-the-art algorithms in experimental comparison.