Paris, France
PhD Student at Université Paris Cité (LIPADE/SIP) × Forvia, working broadly on self-supervised learning (SSL). Previously graduated from ENSAE Paris and earned an MSc in Data Science & AI at Institut Polytechnique de Paris.
Research interests
News
Starting PhD at Forvia × Université Paris Cité (LIPADE) on self-supervised learning.
Officially graduated from Institut Polytechnique de Paris with an MSc in Data Science.
Officially graduated from ENSAE Paris (IP Paris) with a Master's degree (Diplôme d'ingénieur) in Data Science, Statistics and Machine Learning.
Paper accepted at the IAAI workshop @ ECAI 2025 — MetaRAG: Metamorphic Testing for Hallucination Detection in RAG Systems. [link]
Publications
Channdeth Sok, David Luz, Yacine Haddam
IAAI Workshop @ ECAI 2025 · Bologna, Italy (CEUR-WS Vol-4136)
Proposes MetaRAG, a metamorphic testing framework to systematically detect hallucinations in Retrieval-Augmented Generation systems by evaluating consistency under controlled input transformations.
Experience
June 2026 – Present
Industrial PhD (CIFRE) between Forvia's Paris Tech Center and LIPADE, Université Paris Cité, broadly working on self-supervised learning for the automotive sector.
May – Nov 2025
Built MetaRAG for hallucination detection in LLM and RAG systems and DARIA for automated analytical reporting, delivering AI-powered solutions within Forvia's data science team.
Mar – Jul 2024
Developed an end-to-end document-based conversational AI chatbot with advanced retrieval capabilities. Implemented hybrid search combining vector embeddings and keyword matching, experimented with multiple chunking strategies to optimize context retrieval, and added source citation with sentence-level highlighting. Evaluated system performance using LLM-as-a-judge methodology to ensure answer quality and accuracy.
Jun – Sep 2023
Studied state-of-the-art LLMs and built a GPT-3.5/LLaMA-2 system to evaluate résumés against job descriptions, producing ranked candidate shortlists with explainable qualification decisions.
Education
2026 — Present
Université Paris Cité
École Doctorale Informatique, Télécommunications et Électronique de Paris (EDITE — ED 130).
2024 — 2025
Institut Polytechnique de Paris | IP Paris
Relevant courses: Deep learning, NLP, Advanced AI, Big Data, Reinforcement Learning.
2022 — 2025
ENSAE Paris
Relevant courses: Statistics, Machine Learning, Deep Learning, NLP, Bayesian Statistics, Econometrics, Optimal Transport, Macroeconomics, Microeconomics.
2019 — 2022
Institut de Technologie du Cambodge | ITC
Two years of intensive courses in Mathematics, Physics and Computer Science.
Projects
Research Implementation
Reproduced the convergence, regularization path, and debiasing results from Lénaïc Chizat (2024) using OTT-JAX and packaged the workflow in Google Colab for peers.
Delivered annotated notebooks and benchmarks validating annealed Sinkhorn behavior across datasets.
VINCI · Jan–Apr 2025
A model-agnostic retriever package for Python: robust enough to work out of the box on generic use cases, yet easy to customize for domain-specific LLM and RAG pipelines.
Built a flexible retriever module that can be easily integrated into RAG pipelines, improving retrieval relevance and reducing hallucinations across diverse applications.
Data Science Sprint
Constructed composite indicators to score countries on education investment attractiveness, blending macro trends with education KPIs.
Produced a ranked list of countries with actionable insights for policymakers and investors, highlighting key drivers of education investment potential.
Collège de France · 2022–2023
Analyzed DEPP Premier Degré panel data to uncover determinants of mathematics performance gaps among French students.
Identified key socioeconomic and pedagogical factors to inform policy recommendations.
Contact
I’m open to research collaborations and conversations about self-supervised learning, LLM evaluation and RAG — across academia and industry.