Hi Aditya,
I think, 'waver' program in AFNI which takes stimtimes as input and outputs a regressor at TR level always uses a default virtual TR of 0.1 sec for convolution with canonical hrf and only at the end downsamples the convolved signal to the TR specified by the user.
So in your case, you can create the Psychological regressor at Virtual TR level (like 0.1 sec) and also interpolate the time-series from Seed data (after de-convolving with 3dTfitter) to this sampling rate, multiply them at this virtual TR level and convolve with hrf and then finally downsample back to your experimental TR.
Best,
Srikanth.