Abstract: Molecular beam epitaxy (MBE) is an atomically precise method for the synthesis of extremely thin films which may possess unique and desirable functionalities. The epitaxial growth process is typically monitored by reflection high energy electron diffraction (RHEED), presenting information on surface morphology, growth rate, and crystallinity. However, observing and interpreting RHEED patterns is both time intensive and complex. In this work, we are developing an artificial intelligence (AI)-driven pipeline to enable automatic monitoring of the deposition process via real-time RHEED image analysis (one image per second) for targeted materials. Our pipeline utilizes a pre-trained image model that encodes each RHEED pattern image into a feature vector. Changes in the RHEED pattern are detected via two analytics methods: a time series-based changepoint detection method that measures changes in pairwise cosine similarity between feature vectors, and a graph theoretic method that clusters feature vectors by cosine similarity. We implement the open source framework and detect physically meaningful changes in RHEED videos collected from the deposition of epitaxial thin films such as anatase $\ce{TiO2}$ on $\ce{SrTiO3}(001)$. We present the strengths and weaknesses of this approach and its potential use as the basis for on-the-fly feedback control of MBE deposition parameters.