Accessible and Automated City Service Requests: A Machine Learning Approach

Using machine learning, this project will develop an image processing pipeline that takes a user submitted image from Boston’s 311 app and helps the resident submit a corresponding non-emergency city request. By automating this process, this service can be accessible and more intuitive for a wider range of audiences, who may not have English as their first language or be well-versed with using technology, while also consuming less of the city’s resources due to erroneous submissions. Curating a dataset from the previously submitted 311 system photos, I will prepare a large image collection that will then be used to develop a CNN for image classification with the TensorFlow library. Then, I will develop a large language model trained on past service requests that can help draft a service request for the user that is customized to the specific situation at hand. These models will be saved, compressed, then deployed to an accessible API endpoint.