This step, particularly for unresolved or moderately resolved sources is easily automated. The three major operations needed for this step are (1) gridding/de-gridding the visibility data onto a regular grid, (2) Fourier transforming the gridded visibilities and the weights to form the Dirty Image and the PSF, and (3) deconvolve the Dirty Image using one of the standard deconvolution algorithms. Given a PSF, a minimum flux limit or the maximum number of iterations, and an image-plane window encompassing the compact source of interest, this step is already automated using the current deconvolution algorithms. High dynamic range and fidelity images can be made even more easily using the more modern algorithms (e.g. scale sensitive deconvolution algorithms).
Programs already exists with the authors which implements the deconvolution algorithm and can be easily modified to accept the GMRT native data recording format. It is therefore reasonable to assume that this step can be certainly automated.