We present a near real-time rateless coding approach based on the constrained use of a feedback channel. The specific feedback used in our scheme is based on a measure of distance between a received word and the symbols already decoded at the receiver, with successive feedbacks providing accumulating information about the state of the decoder. Our encoder, in turn, optimizes its choice of constituent source symbols in each encoding according to the decoder information and not uniformly at random as is typically done. We provide an upper-bound on the error probability of our approach when used with a maximum-likelihood decoder, and simulation results demonstrate that the coding performance can be tuned according to the feedback budget available.