Database Transformation

Context

Organizations sit on vast amounts of relational data, but turning it into a knowledge graph — the format that powers semantic search, analytics, and more trustworthy AI — has been slow, expert-only work: hand-writing mapping rules in a syntax-heavy language. I built LXS to make it a guided, human-in-the-loop process instead. The user connects a database and loads the relevant ontologies; the system drafts the mapping rules that link each table column to the right ontology class and property, breaks the job into small steps, and shows a plausibility signal on every suggestion so a non-expert can review, accept, or correct it with confidence. It is neuro-symbolic with a retrieval-augmented core: a symbolic engine retrieves the best-fitting ontology snippets and pulls in vocabularies from external terminology services, and the language model drafts from exactly that context — so suggestions stay grounded, transparent, and traceable. In evaluation it produced higher-quality mappings than manual coding, an established editor, and AI working on its own, while cutting the hands-on effort dramatically. Most telling, quality held steady regardless of the user's background, so people new to the semantic web matched expert output — and keeping a person in the loop with visible plausibility checks builds the trust that fully automated pipelines tend to lack.