Improving Gradient-guided Nested Sampling for Posterior Inference