Fifth generation (5G) networks are currently being deployed at millimeter wave (mmWave) bands, beyond 22 GHz. The wireless node density and gigabit-per-second demands of 5G Internet-of Things (IoT) devices are pushing for more spatial reuse and higher frequency bands, which can be achieved by directional beamforming methods. Over the years, researchers have relied on synthetic data and simulation for studying directionality and beamforming, due to the lack and high cost of mmWave hardware. Hence, there is a major need for software-defined radio (SDR) platforms that enable programmable directionality in wireless studies and experimentation. Recently, more affordable and commercially available mmWave radio frequency (RF) front-ends with off-the shelf SDRs have made it possible to set up experimental test-bed platforms for beam alignment studies. In this thesis, we present a low-cost “directional SDR” test-bed that enables convenient programming of mmWave beam directions from a high-level programming language. The test bed design allows modular use of different mmWave antenna systems, including horn and path array antennas. Using a multi-threaded software configuration, the test-bed facilitates real-time access to legacy SDR methods including machine learning (ML) algorithm libraries. With a focus on receiver side Angle-of-Arrival (AoA) detection as a use case, we demonstrate the test-bed’s capabilities in ML-based mmWave beamforming.