Once data has been collected, it is desirable to be able to quickly transform that data into statements about the values and corresponding uncertainties — the “confidence limits” or “credible limits” — of the physical parameters underlying the data.  Traditionally, this problem is treated probabilistically. This process can be time consuming, as enough samples need to be drawn that the distribution of samples converges to the Bayesian posterior distribution of the likelihood in parameter space.  We propose an alternative scheme based on the likelihood ratio test, minimizing a cost function that incentivizes exploration of parameter space points within the credible limit but far from previously explored points.  Testing on toy likelihood functions with realistic non-Gaussian properties, we find that this scheme converges to the same credible limit as the usual Bayesian methods but requires an order of magnitude fewer evaluations of the likelihood function.  Our code is publically available over GitHub.