Level

Level processes raw applicant records through a FastAPI pipeline using Cohere embeddings, UMAP reduction, and DBSCAN clustering to quantify distribution skew. Its front-end allows users to manipulate these clusters, generating a clean, intentional dataset for model training.