LookUpMark/semanticmesh — reverse-engineered prompt

Build me a Python app for data governance that can take business documents like PDF or TXT files plus database schema files in SQL, then automatically figure out how the business concepts relate to tables and columns. I want it to create a Neo4j knowledge graph from that mapping, handle duplicate or similar terms sensibly, and validate what it builds so bad mappings or broken graph queries get corrected before saving.

I also want a question answering flow where I can ask things in plain English like what table stores customer status or how a business term maps to the database, and get a grounded answer with sources from the docs and schema. If the system is not confident or there is not enough evidence, it should say so instead of making something up. For low confidence mapping steps, add a simple human review checkpoint.

Please make it easy to configure with different LLM providers and include a basic way to run ingestion, graph building, and querying end to end. Look up current docs online if you need to.