Beck LaBash

• RES-Q: Evaluating Code-Editing Large Language Model Systems at the Repository Scale, arxiv, 2024
• Associating Underwater Imagery with Geolocation for The Study of Eelgrass Health, NEERS Fall Meeting, 2020
• Classifying Underwater Eelgrass Imagery with Machine Learning, EPA Region 1 Spring Meeting, 2021

• Generative Flow Networks for Neural Network Hyperparameter Tuning: Proposed a general framework to apply Generative Flow Networks to hyperparameter selection. Showed that framework can sample diverse, performant batch size, learning rate, and epoch configurations for training a feed-forward neural network on the Fashion-MNIST dataset.
• A Video Is All You Need: Learning Fine-Grained Running Form Representations from Monocular Sources: Fine-tuned a Dual-stream Spatio-temporal Transformer to learn a dense representation of a subject's running form, given a single input video of them running. Showed that the latent representation learned by the transformer, combined with simple vector classification methods, can be used to identify flaws in form, potentially predicting future injuries.
• CRN2Cal: React app that converts course reference numbers into calendar events, allowing users to easily schedule courses in any calendar app.
• @MassVax: Scraped and analyzed real-time vaccine availability data from CVS pharmacies during the COVID-19 pandemic. Leveraged Twitter API to publicly tweet vaccine availability changes 24/7, gaining ~10,000 followers at peak.

Working as an AI Consultant at Qurrent AI

Working as a Machine Learning Co-op at Notch Technologies

Leading a team at Generate Product Development to build the backend of an NLP-enabled mental health journaling platform.