I am trying to implement k-epsilon model for a complex geometry involving recirculation. Unless I use the inlet turbulence intensity greater than 10% (usually specified as 2-5%), the code is not converging. I wasnt sure, if this would spoil the accuracy of k-epsilon model. Does anybody has any experience with the effect of inlet boundary conditions for k and epsilon.

You should wait till after several hundred times of iteration, it may change to converge. i don't know what is your Reynolds number. if Re is not very high you can set the kinetic energy and turbulence dissipation to 0.1 with k-epsilon model.