Agentic AI Applications with Graphs & Open-Weight Models
Register: https://www.meetup.com/data-science-community-meetup/events/308060973
We're excited to announce our upcoming AI Belgium community meetup featuring two great speakers who will dive deep into cutting-edge AI topics to help you stay up-to-date and drive technology choices for your Artificial Intelligence applications.
Thank you to Neo4J for sponsoring this meetup and the Commons hub for hosting us.
Sponsors allow us to make the meetups free: we are an open non profit tech community, openly sharing experiences from both an industry and/or an academic perspective. This meetup’s sponsor is Neo4J, the graph database and analytics company. Thank you to the Commons hub as well for hosting this event.
AGENDA
6:00 PM: Opening
6:30 PM: Tech talks (more information below)
8:00 PM: Informal networking drink with pizza (thanks Neo4J!)
1. Sofie Van Landeghem - Inside the Black Box: What Open-Weight LLMs Today Can (and Can't) Do
As open-weight large language models continue to close the gap with proprietary APIs, developers and researchers face a new landscape of possibilities and trade-offs. This talk explores the current capabilities of open models, highlighting the strengths and weaknesses of the most popular model families. We'll focus in particular on how to navigate this landscape when developing production-ready NLP applications.
2. Tomaz Bratanic - Beyond Vector Search: How GraphRAG Transforms Information Retrieval
Retrieval-Augmented Generation (RAG) is a widely used technique for grounding LLM outputs in external knowledge. However, standard RAG approaches often struggle with complex information needs, especially when relationships, entities, or broader context matter. GraphRAG extends vanilla RAG by incorporating knowledge graphs to structure the context data used during retrieval. This talk introduces the core concepts behind GraphRAG: how knowledge graphs can represent and organize retrieved information, how they complement vector similarity search, and how they enable more targeted, explainable, and complete retrieval.