A Knowledge Graph for a Century of Expertise

Large language models (LLMs) can finally make sense of unstructured documents, but first-generation retrieval-augmented generation (RAG) often falls short on performance and explainability. Knowledge graphs, which link entities and their relationships, are emerging as a way to give LLMs reliable, explainable context.

A global beauty group had invested for years in structured data, yet more than a century of its history remained buried in around 10 million unstructured documents.

The group needed to:

• Unlock the knowledge locked in its unstructured documents.
• Overcome the performance and explainability limits of its first RAG use cases.
• Link structured and unstructured data so that both experts and LLMs can use it.

Sia built the foundations of the group's unstructured data program, powered by knowledge graphs and LLMs:

• Framework: an end-to-end framework for unstructured data projects, covering skills, process, governance and tooling.
• First use cases: data recovery, claims extraction and formulation knowledge evaluation.
• Graph expertise: end-to-end graph expertise, from ontology and entity extraction to GraphRAG, with technology partners.

The group gained a knowledge layer linking structured and unstructured data for experts and LLMs alike:

• Its first knowledge graphs, accessible through web apps and a GraphRAG chat.
• A framework to run further unstructured data projects.

By combining knowledge graphs, LLMs and a clear data governance framework, Sia helps organizations turn their unstructured knowledge into a strategic asset.