Hi! I am Guilhem Fouilhé, a PhD student at IRIT and member of the MELODI team. I am also part of ANITI, one of the French AI clusters in Toulouse.
AI researcher and engineer finishing a PhD on explainable AI, with a core focus on human-aware AI: designing conversational interfaces and multi-agent systems that collaborate with people in complex decision-making. By combining explainable AI (XAI) with dialogue interfaces over symbolic planners, I build systems where humans can query, explore, and steer decision processes through interactive goal-conflict explanations.
I pair a strong mathematical foundation with end-to-end systems engineering: from low-level GPU inference (vLLM, CUDA) and multi-agent LLM orchestration (OpenAI API, Ollama, OpenClaw) to full-stack interactive applications and empirical human-in-the-loop user studies measuring trust, comprehension, and decision quality.
Outside of research, I enjoy spending time outdoors in wide open spaces near the Pyrenees and travelling to learn about different cultures and ways of life.
Human-aware AI for decision-making: conversational interfaces, interactive XAI, and human-in-the-loop evaluation.
AI systems designed to collaborate with human planners in sequential decision problems — modeling preferences and facilitating iterative co-planning rather than autonomous black-box replacement.
Conversational explanations grounded in user questions. Providing goal-conflict and contrastive explanations so decision-makers can calibrate trust and understand trade-offs.
Multi-agent architectures that mediate between users, domain tools, and symbolic planners on OpenAI and self-hosted backends (Ollama, OpenClaw, vLLM).
Empirical user studies and automated benchmarks, evaluating whether human decision-makers comprehend, trust, and can act on system explanations.
Research and engineering across human-aware AI, conversational decision systems, explainability, and applied mathematics.
IRIT & ANITI, Université de Toulouse
Developing human-aware AI systems for collaborative decision-making: multi-agent LLM architectures, conversational interfaces over automated planning, interactive goal-conflict explanations, and human-in-the-loop user studies. Advisors: Nicholas Asher and Philippe Muller. Funded by the EU project TUPLES.
IRT Saint Exupéry
Supervisors: David Vigouroux and Agustin Picard. Influence functions for global explainability of deep neural networks. Report
CerCo – CNRS (funded by ANITI)
Supervisors: Grégory Faye and Rufin VanRullen. Mathematical models for predictive coding in vision. Theoretical and applied study of stability of predictive coding DNNs (PyTorch). Publication
IMT – Institut de Mathématiques de Toulouse (funded by CIMI)
Supervisors: Grégory Faye and Christophe Besse. Sharp thresholds for bistable reaction–diffusion equations. Theoretical and numerical study of asymptotic behavior. Publication
Université de Toulouse, IRIT
Advisors: Nicholas Asher and Philippe Muller. Expected defense: December 2026.
Université Toulouse III – Paul Sabatier
Valedictorian (major de promotion), with honours. ANITI Graduate School Fellowship (M1 and M2).
Université de Montpellier
First two years as a Mathematics–Physics double major. With honours.
Peer-reviewed papers, preprints, and academic reports. Click to expand abstracts and citations.
We investigate the influential factors governing the trade-off between parametric memory and in-context information during factual recall in Large Language Models.
Accepted at BlackboxNLP 2026 (co-located with EMNLP 2026 in Budapest, Hungary). Proceedings forthcoming.
We formalize and implement an agentic, multi-agent LLM architecture for conversational explanations in planning, and evaluate it with a human-in-the-loop user study.
When automating plan generation for a real-world sequential decision problem, the goal is often not to replace the human planner, but to facilitate an iterative reasoning and elicitation process, where the human’s role is to guide the AI planner according to their preferences and expertise. In this context, explanations that respond to users’ questions are crucial to improve their understanding of potential solutions and increase their trust in the system. To enable natural interaction with such a system, we present a multi-agent Large Language Model (LLM) architecture that is agnostic to the explanation framework and enables user- and context-dependent interactive explanations. We also describe an instantiation of this framework for goal-conflict explanations, which we use to conduct a user study comparing the LLM-powered interaction with a baseline template-based explanation interface.
This is a demo paper about a platform for iterative planning with interactive goal-conflict explanations. My most notable contribution is the design of an agentic multi-agent LLM architecture to present explanations to users through a chat interface, including human-in-the-loop evaluation via online user studies.
When automating plan generation for a real-world sequential decision problem, the objective is often not to replace the human planner, but rather to facilitate an iterative reasoning and elicitation process, where the human’s role is to guide the planner according to their preferences and expertise. In this context, explanations that address users’ questions are crucial to improve their understanding of potential solutions and increase their trust in the system. We present a platform that implements this iterative planning approach and provides explanations to user questions based on conflicting goals and preferences. The platform supports both a classical template-based interface and a multi-agent Large Language Model (LLM) architecture that enables interactive explanations tailored to the user and context. The integration of online user studies allows for the evaluation of the effectiveness of the explanations and the impact of the communication interface.
We formalize and implement an agentic, multi-agent LLM architecture for conversational explanations in planning.
This is my first first-author paper!
When automating plan generation for a real-world sequential decision problem, the goal is often not to replace the human planner, but to facilitate the tedious work. In an iterative pro- cess, the human’s role is to guide the planner according to their preferences and expert experience. Explanations that re- spond to users’ questions are crucial to increase trust in the system and improve understanding of the sample solutions. To enable natural interaction with such a system, we present an explanation framework agnostic architecture for interac- tive natural language explanations that enables user and con- text dependent interactions. We propose conversational inter- faces based on Large Language Models (LLMs) and instan- tiate the explanation framework with goal-conflict explana- tions. As a basis for future evaluation, we provide a tool for domain experts that implements our interactive natural lan- guage explanation architecture.
We investigate how transformer models learn linear functions through in-context learning, challenging common assumptions about their algorithmic capabilities.
This work provides new insights into the limitations of in-context learning by studying a controlled setup with synthetic data.
We explore in-context learning (ICL), a popular paradigm for inference with Large Language Models (LLMs), in a controlled experimental setup using synthetic training data. Using a range of small transformer models trained from scratch, we focus on a mathematical task with simple yet precise prompts: learning a linear function f from a sequence of inputs and their corresponding function values . Our findings challenge the prevailing narrative that transformers adopt algorithmic approaches like linear regression to in-context learn (ICL) a linear function. We observe that all models have “boundary values” that limit generalizability. While we can extend boundary values with training distributions over a wider range, we lose the precision of models trained on distributions with more restricted ranges. Thus, we see a dilemma for ICL at least in some tasks: either models will lack generalizability or precision.
We present a mathematical framework to systematically investigate neural dynamics in a hierarchical perceptual system, showing how stability and wave propagation properties emerge from predictive coding principles.
This paper provides a rigorous mathematical analysis of hierarchical neural networks with predictive coding dynamics, revealing conditions for stability and characterizing wave propagation properties.
Sensory perception (e.g., vision) relies on a hierarchy of cortical areas, in which neural activity propagates in both directions, to convey information not only about sensory inputs but also about cognitive states, expectations and predictions. At the macroscopic scale, neurophysiological experiments have described the corresponding neural signals as both forward and backward-travelling waves, sometimes with characteristic oscillatory signatures. It remains unclear, however, how such activity patterns relate to specific functional properties of the perceptual apparatus. Here, we present a mathematical framework, inspired by neural network models of predictive coding, to systematically investigate neural dynamics in a hierarchical perceptual system. We show that stability of the system can be systematically derived from the values of hyper-parameters controlling the different signals (related to bottom-up inputs, top-down prediction and error correction). Similarly, it is possible to determine in which direction, and at what speed neural activity propagates in the system. Different neural assemblies (reflecting distinct eigenvectors of the connectivity matrices) can simultaneously and independently display different properties in terms of stability, propagation speed or direction. We also derive continuous-limit versions of the system, both in time and in neural space. Finally, we analyze the possible influence of transmission delays between layers, and reveal the emergence of oscillations.
We study the asymptotic behavior of nonlocal bistable reaction-diffusion equations.
In this paper, we study the asymptotic behavior of the solutions of nonlocal bistable reaction-diffusion equations starting from compactly supported initial conditions. Depending on the relationship between the nonlinearity, the interaction kernel and the diffusion coefficient, we show that the solutions can either: propagate, go extinct or remain pinned. We especially focus on the latter regime where solutions are pinned by thoroughly studying discontinuous ground state solutions of the problem for a specific interaction kernel serving as a case study. We also present a detailed numerical analysis of the problem.
We review the application of predictive coding - a popular neuroscience framework - to deep neural networks.
This work was produced for TIR (Travaux d’Initiation `a la Recherche) course in M1 Computer Science of UPS during year 2021/2022.
Predictive Coding is a popular framework in neurosciences for explaining cortical function. In this model, higher-level cortical areas try to predict lower-level neural activity and prediction errors are passed back to higher layers. Deep Neural Networks (DNN) , which use brain-inspired architecture, could be augmented with such a model, providing robustness and a better understanding of spatio-temporal dependencies. We investigate research in this direction and give a quick review on tasks in which Predictive Coding (PC) for DNN has demonstrated its interest, with a strong emphasis on vision-related tasks.
Human-aware AI platforms, conversational decision interfaces, and LLM systems — from GPU training to deployed interactive agents.
Pretraining a billion-parameter state-of-the-art LLM from scratch. HPC GPU training, inference, and evaluation with PyTorch Lightning, Nvidia NeMo, Hugging Face, and custom Triton kernels.
Human-aware AI platform for iterative planning with conversational goal-conflict explanations. Multi-agent LLM architecture bridging users and symbolic planners, evaluated with human-in-the-loop user studies.
Interactive visualization for temporally annotated data (TimeML / TimeBank / TempEval).
Academic project at IRIT (supervised by Chloé Braud): literature review and predictive modelling of linguistic markers associated with dementia.
Visualization tools for a Pl@ntNet dataset at the Institut Montpelliérain Alexandre Grothendieck.
Human-aware AI engineering, conversational systems, LLM serving, and human-in-the-loop evaluation — from GPU inference to interactive decision platforms.
Conversational interfaces and multi-agent LLM systems for decision-making and planning on proprietary (OpenAI) and self-hosted backends (Ollama, OpenClaw, vLLM).
Goal-conflict explanations, contrastive reasoning, user studies, and empirical evaluation of trust, usability, and cognitive alignment.
GPU access and scheduling, vLLM inference, and throughput / latency optimization for production model serving.
Containerized deployments, CI/CD pipelines, AWS cloud infrastructure, and interactive web tools (Angular, FastAPI, Node.js).
Lectures, tutorials, and lab sessions for undergraduate (L1–L3) and graduate (M1) students at Université Toulouse III – Paul Sabatier and UPSSITECH engineering school.
I co-authored a national online module, “Identifier les possibilités d'utilisation des IA génératives textuelles”, for the French Université Numérique. This module, accessible to all students in France, explores the potential uses of generative AI in academic settings. My role was to explain how ChatGPT and similar models function.
Open to applied AI, LLM engineering, and research roles from late 2026 — especially agentic systems, LLM serving, evaluation, and MLOps. The easiest way to reach me is by email.
Academic: guilhem.fouilhe-lafforgue@irit.fr
