User signals · prediction
Based in France · Open to remote opportunities
Product & behavioral data science
I use experiments and predictive models to answer product questions.
My work combines behavioral and product data, statistical analysis and machine learning. I evaluate what a signal predicts, where it fails to generalize, and what the result supports a team doing next.
PhDCognitive Science
5+ yearsApplied AI & research
R&DBehavior · retrieval · research
Offline → ProdFrom benchmark to deployed service
Scientific ReportsPublication · Nature Portfolio
Core expertise
Personalization · search
Text · image · recommendation
Serving · monitoring · load
Student supervision · project leadership
01 — Selected evidence
Predictive signals, evaluation and decisions.
Published research and professional case studies, with the contribution, outcome and limits of each piece of work made explicit.
01Behavioral modeling
View all six projects ↗Predictive segmentation: why website transfer changed the decision
Predictive signal on known sites; insufficient transfer to another site led to stopping before productionization.
Feasibility study stopped before productionization.↗02Ranking personalizationPersonalizing search through user affinities
Affinities outperform context alone across offline NDCG, Hit Rate and MRR; online uplift remains to be measured.
Professional prototype evaluated offline.↗03Behavioral artificial intelligenceCan the way we look really reveal personality?
Some gaze characteristics vary with personality; the most credible are those that reappear across tasks, content and moments of exploration.
Published research · Scientific Reports, 2024↗04Multimodal UX evaluationGenerative AI for UX evaluation: comparing model and human judgments
100 screenshots × 6 dimensions; Pearson 0.343–0.535 by dimension, without a calibration claim.
Offline experiment; saved results and reproduced correlations.↗02 — One throughline
Behavior → AI → Product → Impact.
One line of reasoning connects my projects: start with a product question, choose useful signals, build a model and make its output understandable.
CursorAI
Movement, pauses, clicks and scrolling
AttentionAI · DeepUX
Likely gaze and interface perception
SearchAI
Meaning, context and visual resemblance
TalentMap
Mentioned skills, working preferences and nearby profiles
Synthetic UX
Hypotheses and questions to validate in the field
03 — ML products at AB Tasty
From research protocol to deployed system.
Five case studies show how I helped turn hypotheses into products and innovations across segmentation, search, recommendation and AI-assisted UX.
01Behavioral modeling
View all case studies ↗Predictive segmentation: why website transfer changed the decision
Predictive signal on known sites; insufficient transfer to another site led to stopping before productionization.↗02Multimodal retrievalMultimodal search and recommendation: from image to production pipeline
Task-specific model selection and production implementation of an incremental text-image pipeline, with monitoring and load testing.↗03Ranking personalizationPersonalizing search through user affinities
Affinities outperform context alone across offline NDCG, Hit Rate and MRR; online uplift remains to be measured.↗04Multimodal UX evaluationGenerative AI for UX evaluation: comparing model and human judgments
100 screenshots × 6 dimensions; Pearson 0.343–0.535 by dimension, without a calibration claim.↗05Retrieval model evaluationSelect an embedding model by task, not by popularity
OpenAI leads text-to-product retrieval; FashionCLIP improves visual NDCG@10 by 4.8 and 4.0 points across two public datasets.↗04 — Experience
Applied research, product and production.
View full résumé ↗Research Engineer · AI R&D
AB TastyFrom benchmarks to search, recommendation and personalization systems integrated into high-traffic products.
- Multimodal pipeline shipped for search and recommendation
- Predictive segmentation and cross-site transfer validation
- Evaluation protocols connecting ML performance, product use and limits
Research Engineer · PhD Research Scientist
Dotaki / AB Tasty · Université Paris CitéAn applied PhD connecting personality, visual attention, digital interactions and product personalization.
- Behavioral metrics derived from gaze and mouse movements
- Research findings published in Scientific Reports — Nature Portfolio
- Student supervision and research project leadership
- From cognitive models to signals integrated into a product
05 — Published research
SCIENTIFIC REPORTS · NATURE PORTFOLIO · 2024Cognitive Science explains what my products are designed to measure.
My PhD connects personality, visual attention and digital interactions. It provides the experimental framework for separating a useful signal from a merely technical correlation.
View my publications ↗06 — Compact stack
Tools in service of the product.
AI / ML
PyTorch · TensorFlow · scikit-learn · XGBoost · embeddings
Product engineering
Python · FastAPI · Docker · GCP
Data & retrieval
ClickHouse · Meilisearch · GCS · CLIP · SBERT
07 — Contact
Let’s build products that understand their users.
Open to Applied Scientist, AI Research Engineer and AI Product Builder roles.
Start a conversation ↗