With a PhD in Fundamental Sciences and dual expertise in Computer Science, I help laboratories and industries navigate the next generation of AI. I specialize in building autonomous agentic systems and advanced RAG architectures that turn complex data into actionable insights.
Whether it's optimizing industrial processes or automating scientific monitoring, I design custom, sovereign solutions that prioritize data privacy and technical rigor. For French speaker, feel free to visit my website with my portfolio and articles !
- Advanced RAG & Strategic Monitoring: Building semantic search engines and automated watchlists (competitive intelligence, internal docs) with local inference to ensure data sovereignty.
- Agentic Ecosystems: Developing autonomous agents capable of managing complex workflows—from automated experimental analysis to deep synthesis of academic corpuses.
- AI for Deep Science: Leveraging cutting-edge architectures (GNNs, PINNs) and LLM reasoning to model complex systems and facilitate data-driven decision-making.
- Local Infrastructure: Full-stack development on high-performance local setups (Multi-GPU) for offline training and execution (No cloud data leaks).
| Category | Tools & Technologies |
|---|---|
| GenAI & Agents | LangGraph, LangChain, n8n, Ollama, Unsloth, ChromaDB |
| Data Science | Python (PyTorch, Scikit-Learn, Optuna), FastAPI, RDKit, PubChemPy |
| Interfaces | Streamlit, Plotly, Node.js, Electron |
-
Bioactivity prediction using Graph Neural Networks on molecular structures. Trained on ChEMBL data, this model predicts whether a compound is active against specific biological targets.
👉 See the presentation of the project (in French!) -
Prediction of photochemical properties (HOMO-LUMO gap, orbital energies, redox potentials) using Graph Neural Networks. Leverages quantum chemistry datasets (QM9, PC9, Transition1x, Harvard OPV) to train models capable of accelerating photocatalyst design and organic solar cell discovery.
-
AI-powered automated monitoring ecosystem designed for scientific research. Orchestrates AI agents, advanced RAG, and local LLMs within a user interface that allows researchers to interact with their data and watchlists. A deep agentic ecosystem automates complex tasks to significantly accelerate daily research workflows.
-
Implementation of GraphRAG pipelines inspired by Microsoft’s framework and rebuilt in Python, leveraging local LLMs via Ollama for large-scale corpus processing. Demonstrated on 25 academic publications and 25 theses, using local models for inference and automated evaluation through LLM-as-a-Judge scoring. Includes a graph visualization interface for manual data exploration.
-
Development of an expert model specialized in agrochemical law through fine-tuning open-weight models. The goal is to consistently outperform generalist LLMs on compliance, authorization, and substance status queries within French and European regulatory frameworks.
If you're working on projects related to sustainable chemistry, clean energy, or molecular innovation, feel free to reach out. I'd be happy to discuss with like-minded scientists!