This summer, I had a 12 week internship at a photonics company called Thorlabs, doing research in machine learning for one of their products. It was an incredibly rewarding experience and I learned a great deal about independently working on an end-to-end project. My research involved collected images from a fiber optic cleaving device and developing a machine learning pipeline to first classify successful vs. unsuccessful images, and also predict the optimal machine parameters for the cleave device. Images were taken manually over a two week period at the start of the internship, with over 400 cleave images taken. Metadata was then downloaded from a handset connected to the cleave machine, including parameters for the tensions used during the cleave, the resulting angle of the end face, the fiber type, and whether the cleaved fiber contained an imperfections that would lead to signal loss. After developing a pipeline for querying the handset and creating a Pandas DataFrame for each dataset, I began constructing my machine learning models. The first step was to create a classification model using a Convolutional Neural Network, specifically with the TensorFlow and Keras Python libraries. After experimenting with different architectures, I tuned the model using data augmentation and hyperparameter optimization to improve generalization. The final classifier achieved an accuracy of 94% in distinguishing successful vs. unsuccessful cleaves. Next, I developed a regression model to predict the optimal machine parameters given a set of cleave images. This step was more challenging due to the limited dataset size and variability in fiber types, but I implemented regularization techniques and cross-validation to maximize performance. I ultimately settled on using the XGBoost library as opposed to a neural network due to the tradeoff in computation time. Although I had difficulty in achieving the desired absolute tension values, I found that the model performed well in identifying the correct direction of tension change, which turned out to be very useful in production. Overall, this internship gave me experience in data collection, pipeline design, deep learning model development, and hardware integration, all within the context of a real-world photonics application.