ESC

Experience

Work Experience

Natural Language Based Secure Federated Query Building
Inria, Wimmics team · Postdoctoral Researcher
June 2026 – June 2029

  • The grounding of LLM in the Knowledge Graph (GraphRAG) helps reduce hallucination in the generated answer.
  • Develop a federated query builder and end-user interface, coupling generative models with linked data for natural language interaction.

Applied Research and Development Engineer
JusMundi · Permanent
November 2024 – May 2026

  • The prior RAG system relied on hard-coded legal rules, reducing flexibility and increasing maintenance. By building a knowledge-graph-based retrieval system using a domain ontology and RDB-to-RDF mappings, we replaced these rules, achieving comparable performance through graph-based retrieval and embeddings.
  • General-purpose embedding models struggled with legal jargon and limited domain data availability. We synthetically generated queries from a large corpus of legal documents and used the resulting query-document pairs to fine-tune a long-context embedding model, achieving a 30%pt recall improvement for the Agentic RAG system.

Query the Web as a Knowledge Graph
Nantes University (LS2N) · Doctoral Researcher
September 2021 – November 2024

  • Can we trust LLMs to extract facts from the Web? We proposed a new evaluation pipeline that carefully uses Schema.org ontology and documentation to verify LLMs' output in the task of extracting Knowledge Graphs from Web pages.
  • Can we still execute queries when the number of Knowledge Graphs increases? First, we created a new benchmark called FedShop that tests the scalability of existing engines in terms of the number of sources. Second, we created a new SPARQL federation engine that finds better source assignment than existing engines.

E-mail personalisation through data
Visage.jobs · Internship
February 2021 – July 2021

  • Using text similarity metrics such as Jaccard or Cosine, we learned that most users use template without personalization.
  • Using text classification methods, we tagged sentences email to extract user writing patterns.
  • Using Association Rule mining, we explain the factors that drive candidate engagement.

Attention-based abstractive summarisation on Yale courses
Nantes University · Internship
January 2020 – May 2020

  • Proposed improvements for ROUGE, METEOR and human evaluation on educational content.

Data visualisation on blockchain network
Cogarius · Internship
April 2019 – June 2019

  • Visualised smart-contracts events on Ethereum; prototype web-application and graph database.

Data visualisation on traffic network
Beuth University of Applied Science · Internship
April 2017 – June 2017

  • Data visualisation on traffic network in Dresden (geo + time).

Education

PhD Computer Science, Semantic Web
Nantes University
Sept 2021 – November 2024

MSc Computer Science, Data Science
Nantes University
Sept 2019 – July 2021

Skills

Large Language Models

  • Framework: LangChain, llama-cpp-python, instructor
  • Platform: Langfuse, Scorecard.io

Database

  • MySQL, PostgresSQL, MongoDB, Spark.
  • GraphDB, Neo4J, Virtuoso, Jena

Development

  • Linux, Bash, Shell script, Docker, CI/CD

Semantic Web

  • Protégé, Virtuoso, Jena, RDF4J
  • Ontop (ODBA), Ontotext GraphDB

Cloud

  • Google Cloud Platform, Azure (ML Studio)

Machine Learning

  • SentenceTransformers, HuggingFace, Pytorch

Certificates

IBM Data Science

  • Date Issued: September 07, 2024
  • Credential: VZ6JZRK7L6XG

IBM AI Engineering

  • Date Issued: July 6, 2024
  • Credential: LV95FKMTPGHB

Languages

English (Professional) · French (Bilingual) · Vietnamese (Native)