Quilt: Custom UIs for Linking Unstructured Documents to Structured Datasets

Practitioners working with large document dumps struggle to identify and record connections between documents and structured datasets. For example, how should journalists building a database of police misconduct incidents connect police report documents to their database of officers? The current approach is either to (i) switch between reading entire documents and querying databases, which is a tedious manual process, or (ii) build a custom linking UI to help. These linking UIs are typically built from scratch using vanilla web programming, despite having similar structures. We introduce Quilt, a framework for generating custom UIs for linking between documents and databases. Quilt’s API lets programmers provide domain knowledge—e.g., what counts as evidence that a particular database row is relevant to a document. From this information, Quilt generates the UI automatically. We explore Quilt’s expressivity via four case studies and find it handles a diverse range of documents and databases.