Session Track: AI Engineering
Session Time:
Session description
Traditional RAG systems often struggle with complex, multi-hop queries that require contextual understanding and structured exploration. In this talk, we’ll present INRAExplorer, a generative AI system powered by a multitool agent architecture and a Neo4j knowledge graph built from open-access scientific publications by INRAE (France’s National Research Institute for Agriculture, Food and Environment). Unlike typical RAG systems, INRAExplorer leverages Neo4j and other custom tools to enable exhaustive, iterative exploration of scientific data. The agent dynamically decides what tool to use as well as when and how to query the graph—retrieving all papers by a specific author, tracing collaboration networks, or answering multifaceted domain-specific questions. This talk will walk through the architecture, the knowledge graph construction process, and live examples showing how combining LLM agents with Neo4j unlocks truly interactive scientific discovery. Join this session to see how agentic RAG systems can turn domain-specific knowledge graphs into powerful tools for reasoning, research, and decision making.
Responsible AI Domain Leader, Ekimetrics
Annabelle Blangero is a PhD neuroscientist and senior manager in data science and responsible AI at Ekimetrics. She leads innovative projects at the intersection of generative AI and scientific research, collaborating with both public and private sector partners. Annabelle is a frequent speaker on such topics as frugal AI, bias mitigation, and interactive systems like CLAIR.bot, designed to make AI reasoning more explainable. She is passionate about making complex information more accessible and actionable, and about ensuring that AI systems are transparent, trustworthy, and grounded in deep domain expertise.
Data Scientist, Ekimetrics
As the head of the NLP research team at Ekimetrics, Jean Lelong specializes in cutting-edge retrieval systems and agentic AI, harnessing the power of Large Language Models to solve real-world challenges. A self-taught programmer, Jean brings a pragmatic and results-driven approach to his work, and is also passionate about chess programming.