This talk is divided into two parts. In the first half of the talk, I will discuss the generation of localized as well as non-local features (i.e. characteristic and repeated patterns that extend over a wide range of scales) in the inflationary scalar power spectrum that lead to a better fit to the CMB data than the conventional power law spectrum. In the second half, I will discuss the non-Gaussianities that arise in such scenarios, and how they compare with the current observational limits on the parameter fNL.