This paper presents and demonstrates the design of a filter-less transmitter architecture with digitally assisted harmonic cancellation. A neural network is used to model the harmonics as well as IMD for digital predistortion applications. This neural network-based harmonic modeling does not require any reference signal to be injected at the input of PA, thereby reducing the complexity in characterizing the harmonics. An in-house 10 W PA operating from VHF to L-band has been designed and characterized for its harmonic suppression. This PA is used along with agile RF transceiver AD9361 from Analog Device and Xilinx embedding platform using Zynq ZC-706 system-on-chip for implementing the entire transmitter. The receiver of AD9361 captures the nonlinearity of the PA in terms of harmonics as well as intermodulation distortion (IMD) components for modeling and predistortion. The proposed architecture can handle all types of distortions due to hardware as well as PA nonlinearity. Besides, it is also able to cancel the harmonics using a harmonic injection in the feed-forward configuration. This transmitter architecture has the advantages of being low cost, filter-less, wideband, frequency agile, reconfigurable and less bulky compared to the conventional scheme. The proposed scheme is demonstrated to transmit 5 MHz LTE signal at different frequencies over the range of 100 MHz to 400 MHz. In such a case, the second and third harmonics appear over the frequency range from 200 MHz to 1.2 GHz, which are within the amplification range of PA, yet they are suppressed without using any filter at the output. More than −40 dBc harmonic rejection is achieved over the entire operating range of this filter-less transmitter. The adjacent channel leakage ratio (ACLR) is always better than −45 dBc after applying digital predistortion.