Julien V. Baudru

Seminar on Artificial Intelligence and Automotive Applications

In this seminar, delivered at IRAM (IRAM Enseignement de Promotion Sociale), I introduced the fundamental concepts of AI in a chronological manner, then explained how it is applied in both traditional and autonomous vehicles, as well as in automotive-related professions. I also presented the objectives of my PhD and discussed the general limitations and challenges of AI.

Presentation at the Belgian Senate

I had the honour of being invited to the Belgian Senate to present my PhD research at an information committee meeting on carpooling. It was an interesting opportunity to explain my work to a non-research audience and contribute to the policy discussion on sustainable mobility solutions.

Optimizing Carpool Service Problem

I recently presented a seminar at LORIA (Nancy) on the Carpooling Service Problem (CSP) and related optimization topics explored in my PhD. The talk covered approaches for user matching, shared routing, and the use of exact and metaheuristic algorithms to address real-world mobility challenges.

How Does AI Work? — Demystifying Artificial Intelligence

Thanks to CEVORA/CEFORA I hosted a live webinar with over 3,000 attendees to explore the fundamentals of artificial intelligence in an accessible way. The session covered how AI learns, the role of data, and the power behind neural networks and generative AI. I also explained key approaches like Machine Learning and Deep Learning, showing how raw data is transformed into intelligent solutions that impact our daily lives.

L'intelligence artificielle en pratique avec Python - 4e édition

I helped write a new chapter for Hugues Bersini's book “Artificial Intelligence in Practice with Python” on the implementation of a simple RAG for the Paris Olympics, to be published in 2025/26.

Early evidence of how LLMs outperform traditional systems on OCR/HTR tasks for historical records

Seorin Kim and I explore the ability of two LLMs -- GPT-4o and Claude Sonnet 3.5 -- to transcribe historical handwritten documents in a tabular format and compare their performance to traditional OCR/HTR systems. Considering the tabular form of the data, two types of experiments are executed: one where the images are split line by line and the other where the entire scan is used as input.

RAG Workshop: Retrieval-Augmented Generation

A workshop organized at HELHa Tournai for computer science students, focusing on Retrieval-Augmented Generation (RAG). This advanced AI technique aims to improve the accuracy and reliability of text generated by Large Language Models (LLMs) by incorporating external knowledge sources.

Institution: Haute École Louvain en Hainaut (HELHa) - Tournai

The Comparative Analysis of Car-Pooling Algorithms for Ride-Sharing Systems

In this paper, we present a comparative analysis of five methods for constructing ride-sharing pools of users, focusing on their efficiency in terms of execution time, the percentage of user requests fulfilled, the distance of the detour made by the driver and the waiting time of the passenger.

Award-Winning Collaboration with Procter & Gamble

IRIDIA and LISA laboratories at ULB received the "Connect + Develop Partner Excellence 2024" award from P&G. Our collaboration focused on implementing Retrieval-Augmented Generation (RAG) systems for the R&D chemical department, optimizing research processes and enhancing knowledge management through advanced AI techniques.

Heuristic Optimal Meeting Point Algorithm for Car-Sharing in Large Multimodal Road Networks

This article introduces a new version of the car-pooling problem (CPP). This involves defining rendezvous or meeting point in such a way that the travel times of the users are fair, this problem shares similarities with the problem of finding the optimal meeting point (OMP) in a graph.

IRIDyOM: Exposing Musical Expectation as an Interactive Creative Space

IRIDyOM is a real-time Max for Live tool that integrates IDyOM-based musical expectation modeling directly into Ableton Live through an interactive, no-code interface. It enables composers and researchers to explore, manipulate, and generate music using listener-relative predictions of surprise and uncertainty within a DAW workflow.

From Theory to Application: Using LLM, RAG, and Embeddings for Real-world Use Cases

The FARI Test and Experience Center in Brussels hosted an AI Happy Hour event. Lluc Bono and I, from IRIDIA (ULB), presented implementations of Large Language Models (LLMs), RAG systems, and embeddings in real-world scenarios. Our presentation demonstrated how to provide the right context to these models for accurate information delivery, outlined integration steps, and discussed methods to reduce errors while improving search capabilities through advanced AI techniques.

Optimization of Carpool Service Problem via Transfer Points Genetic-Based Algorithms

The Carpool Service Problem (CSP) involves matching passengers to drivers while optimizing routing under capacity and detour constraints. We study a variant that incorporates Transfer Points (TPs), allowing passengers to walk to nearby pickup and drop-off locations , thereby increasing matching flexibility.

Linking Network Topology and Mobility Function: Toward Transferable Mobility Agents Across Cities

This work investigates the relationship between network topology and mobility function, proposing a framework for building transferable mobility agents capable of generalizing across different urban environments. The study, presented at BNAIC/BeNeLearn 2025, combines theoretical and empirical analysis of cross-city mobility patterns.

Predicting Agent Trajectory in Urban Road Networks

This paper studies human mobility prediction on Manhattan road networks using synthetic urban trajectories generated with GMMM. It compares BiLSTM, Transformer, and GAT-LSTM models through next-node and full-path forecasting tasks under clean and noisy conditions.

Modeling and Predicting Agent Trajectory in Urban Road Networks

This work addresses the problem of modeling and forecasting agent trajectories in urban road networks, focusing on predictive approaches that leverage structural properties of road graphs to improve accuracy and generalization across mobility scenarios.

Can LLMs outperform classical OCR/HTR pipelines? - Analysis of LLMs' capabilities and limitations in transcribing old handwritten historical records

This study evaluates whether large language models can outperform traditional OCR/HTR systems for transcribing historical handwritten documents. It analyzes their strengths and limitations on noisy, degraded, and stylistically diverse archival sources.

Heuristic Optimal Meeting Point Algorithm for Car-Sharing in Large Multi-modal Road Networks

This work introduces a heuristic approach for identifying optimal meeting points in car-sharing systems operating on large multi-modal road networks. The method focuses on scalability and efficiency for real-time shared mobility coordination.

Lyric Networks Unveiled: A Novel Approach to Analyzing The Beatles' Songs Using Text Embeddings and Network Analysis

This work presents a network-based analysis of The Beatles’ discography using text embeddings and graph methods to uncover thematic structures, lyrical similarities, and relational patterns across songs.

AI Detection of Illegal Animal Trade Online

A collaborative project between VUB and ULB researchers through the FARI Institute to detect illegal animal trade online using artificial intelligence. The system analyzes photos and text from online advertisements to identify illegal practices and has processed over 33,500 ads since January 2022.