Al Barrentine

Al Barrentine is a professional with a diverse background and a strong interest in math, systems, and society. His work primarily revolves around text analysis, topic models, location/geo data, and natural language processing using real-world public datasets like Wikipedia, Twitter, and OpenStreetMap. He focuses on building impactful systems rather than pure research, although his projects involve significant original research.

In the open-source community, Al Barrentine is renowned as the author of the widely used international street address parser libpostal. Developed in C for versatility, libpostal utilizes sequence models to parse complex addresses in multiple languages around the world, drawing from a training set of over a billion examples sourced from OpenStreetMap and OpenAddresses. The entire pipeline for generating training data is open source, making it one of the most extensive labeled sequence training sets available to researchers.

Al Barrentine's expertise extends to collaborative filtering, matrix and tensor factorization, recommenders, network analysis, information retrieval, language modeling, supervised classification and regression, and learning to rank algorithms. He often employs semi-supervised learning techniques to handle partially labeled or mislabeled datasets effectively.

Known for his focus on building learning systems with practical impact, Al Barrentine emphasizes sparse implementations to enable models to run entirely in-memory. Additionally, he leverages his expertise in database design and distributed systems to store and update sufficient statistics for online retrieval, facilitating accurate predictions.

Al Barrentine's educational background includes a Bachelor of Science degree in Social Entrepreneurship, Management & Organizational Behavior, and International Business from NYU Stern School of Business. Throughout his career, he has held key roles at organizations like Florida Rights Restoration Coalition, Second Chances Florida, Mapzen, Paperless Post, SocialFlow, Independent, Jumo, and Lime Company.