Running a semantic embedding model in the browser to make a USWDS combo box filter by meaning instead of substring—hybrid dense/sparse ranking, reciprocal rank fusion, and an honest accounting of what 15 MB of client-side inference actually costs.
A tool that reads dense government regulatory documents—OMB circulars, FISMA, FedRAMP, Section 508, 21st Century IDEA—and uses an LLM to produce plain-language summaries, practical compliance actions, and role-based perspectives for the people who need to understand them.
Knowing which USWDS component to use is half the challenge. The bigger question is how to implement that component in Drupal—which module to install, which paragraph type exists, which Twig template to override.
A structured knowledge base with 146 USWDS components, faceted search indexes, and embedded adaptation guidance—designed so AI coding agents can find the right component and use it correctly.
An MCP server that fetches, parses, and embeds Drupal documentation into a vector database—so coding agents can search, cite, and reason about Drupal APIs instead of guessing.