Research in Artificial Intelligence.
Bridging the gap between foundational AI research and large-scale industrial deployment — I design safe, ethical, compliant and autonomous agentic systems that solve complex real-world challenges.
Modeling Human Behavior and Language for Smarter, Agentic AI Interactions
My research bridges cutting-edge academic theory and heavy-duty industrial demands. Across my career, the unifying thread has been designing systems capable of agentic behavior—from early rule-based architectures exploring the nature of intelligence, to today’s LLM-driven workflows solving real-world challenges.
Thematic Pillars
Pillar 1: Applied NLP, Generative AI & Agentic Engineering (Present & Future)
Context: DSIN of EDF Commerce
My current work focuses on the practical application of advanced NLP and Generative AI within large-scale industrial environments. Building on my background in autonomous dialogue management, I am actively exploring Agentic Engineering to design long-running agents for internal DSIN operations and broad B2B/B2C use cases. A critical component of this work is ensuring these systems remain responsible, safe, and strictly aligned with emerging regulations (GDPR, AI Act).
Key Contributions & Focus Areas:
- Agentic AI & Long-Running Workflows: Designing autonomous systems that manage state, execute multi-step reasoning, and maintain human-agent alignment across prolonged B2B and B2C interactions safely and predictably.
- Generative AI as Reasoning Engines: Leveraging Large Language Models (LLMs) to synthesize synthetic email databases, classify complex customer feedback, and drive automated workflows.
- Data Privacy & Compliance (GDPR): Developing robust, automatic de-identification pipelines for unstructured text to protect user privacy and ensure regulatory compliance in industrial datasets.
- Semantic Exploration: Designing tools to segment dialogue phases in call center transcripts and demystify vast amounts of customer relation data.
Key Publications:
- Génération d’une base de courriers électroniques synthétiques par des grands modèles de langue dans le domaine de la relation client (APIA 2025)
- Désidentification automatique de contenu non structuré (2025)
- Utilisation de LLMs pour la classification d’avis client et comparaison avec une approche classique basée sur CamemBERT (APIA 2024)
Pillar 2: Social Dynamics & Human-Agent Alignment (2014–2017)
Context: Postdoctoral Research (ARIA-VALUSPA at ISIR/CNRS & JOKER at LIMSI-CNRS)
As AI systems become more autonomous, their ability to seamlessly align with human users becomes critical to building systems that genuinely benefit people. My postdoctoral research focused on endowing virtual agents and social robots with advanced socio-affective strategies to maintain user trust and engagement over long-term interactions.
Key Contributions:
- Verbal Alignment & Engagement: Developed quantitative measures and socio-affective strategies to foster user engagement by implementing verbal alignment in human-agent dyadic dialogues.
- Humor & Empathy in Robotics: Contributed to the JOKER project, designing multimodal dialogue systems for robots capable of adapting to human behavioral cues through humor and empathy.
- Open-Domain Conversational Agents: Developed automatic, unsupervised conversation authoring systems utilizing recurrent surface text patterns from large-scale movie subtitle corpora.
Key Resources & Publications:
- Code: Dialign (GitHub) - Automatic measures to characterize verbal alignment.
- Code: GSTLIB (GitHub) - Generalized Suffix Tree Library for pattern extraction.
- Towards alignment strategies in human-agent interactions based on measures of lexical repetitions (Language Resources & Evaluation, 2021)
- Utterance Retrieval based on Recurrent Surface Text Patterns (ECIR 2017)
Pillar 3: Formal Modeling of Task-Oriented Dialogue (2010–2014)
Context: PhD Research (INSA de Rouen & LIFL)
My early research was driven by a desire to understand the nature of intelligence through the lens of human communication. This work laid the groundwork for autonomous, goal-oriented agent behavior by addressing the difficulty of designing systems that can safely navigate multi-step interactions.
Key Contributions:
- Dialogue Games Framework: Proposed a data-driven methodology to extract recurrent interaction patterns from Human-Human dialogues.
- Social Commitments: Designed architectures that allowed agents to manage complex conversational states and autonomously plan their next communicative actions using theoretical frameworks based on social commitments.
- DOGMA Architecture: Implemented this interaction model into DOGMA, an independent module that regulates an agent’s dialogic capacities.
Key Resources & Publications:
- Software: DOGMA - Dialogue Game Manager
- Thesis: Modèle de comportement communicatif conventionnel pour un agent en interaction avec des humains (2014).
- A Conventional Dialogue Model based on Dialogue Patterns (IJAIT 2017).
Research Service
Program Committees
- APIA 2024-2026 - French National Conference on Practical Applications of AI
Reviewing Activity
- APIA 2026 (2 papers)
- APIA 2025 (2 papers)
- APIA 2024 (2 papers)
- ICMI 2020 (3 papers)
- SIGDIAL 2016 (1 paper)
- Interspeech 2016 (1 paper)
- LREC 2016 Workshop ETHI-CA² (3 papers)
- HCI International 2016 (1 paper, Humor in Ambient and Pervasive Interactions session)
- WACAI Workshop 2014 (1 paper)
- TSI Journal 2012 (1 article)
Teaching
Between 2010 and 2016, I taught computer science, software architecture, and mentored students across several French universities. See the full teaching record.
Collaborators
I collaborate with researchers from institutions across France and abroad. See the full list of collaborators.