Enterprise Knowledge Graph
Agentic pipelines that turn scattered public data on European enterprises into a graph you can actually query.
The problem wasn’t finding data about European companies. It was that the data arrived in a dozen shapes from a dozen places, and none of it agreed on what counted as the same entity.
The extraction layer is agentic rather than a fixed pipeline: LLM agents with tool access decide what to pull and how to normalize it, which matters because every source has its own idea of structure. NLP models handle entity resolution, and deep text analysis takes the classification passes too ambiguous for rules. Output lands in a graph store — JanusGraph, later AWS Neptune — alongside vector indexes so the same data is reachable by traversal or by semantic search.
The interesting engineering was in the failure modes. Agents that decide things also decide things wrongly, so most of the work went into constraining what an agent could conclude and making its output checkable.
innoscripta SE · 2021–present