KG Construction Using LLMs

Session Track: AI Engineering

Session Time: 08:00 am - 08:30 am November 6

Session description

Yassir will introduce iText2KG, an open-source framework for incrementally constructing knowledge graphs from unstructured documents using LLMs. He will explain how iText2KG tackles this challenge through a modular pipeline that includes document distillation, incremental entity and relation extraction, and seamless integration with Neo4j for visualization and querying. You will learn how to leverage this pipeline to construct and update knowledge graphs in real time using your textual data. Yassir will also demonstrate techniques for minimizing LLM hallucinations by optimizing the structure and granularity of information passed to the model during graph construction. By the end of the session, you will gain practical skills to apply iText2KG across diverse real-world scenarios—from academic publications and CV parsing to building knowledge graphs from company websites, enabling you to generate dynamic, accurate, and queryable graphs from unstructured text sources.

Speaker

Yassir Lairgi

AI R&D Engineer, Auvalie Innovation

Yassir Lairgi, an AI R&D Engineer at Auvalie Innovation, generates actionable insights for business stakeholders. His journey in AI research and development is strengthened by a strong academic foundation as a current Ph.D. candidate at INSA Lyon, specializing in detecting weak signals in knowledge graph-oriented data, an approach that utilizes LLMs and DGNNs. Yassir's core competencies lie in NLP, LLMs, knowledge graph construction, and graph learning. These skills have been sharpened through the development of iText2KG, an open-source and SOTA algorithm that enables zero-shot KG construction while ensuring consistency within the graph. Beyond academic research, he has applied these technologies in real-world contexts, empowering business leaders by leveraging large-scale data scraping combined with KGs to drive advanced business intelligence solutions.