NVIDIA/auto-ontology ? reverse-engineered prompt

Build me a web app for exploring business data with an AI assistant that can answer questions from connected databases and explain where the answer came from.

I want a polished dashboard with sign in, a chat experience for asking natural language questions, a data catalog view, and settings for managing database connections and model configuration. It should keep a semantic layer or ontology in Postgres, use it to ground terms like metrics and entities, and show provenance or lineage for answers so users can trust the results. Include a backend API, an ingestion flow for tabular data, and a frontend that feels like a modern internal tool. Make it easy to run locally with Docker Compose, and wire in the auth and role based access so only allowed users can see the right data. If you need current setup details or best practices, look up the latest docs online and follow them.