THOMAS LE BRAS, PhD
Research Engineer — 5+ years of experience designing and deploying scalable AI systems for high-traffic products, leveraging research-driven methods across user behavior modeling, data and UX.
“From data and research to product”
I use research as a practical tool to guide product decisions — from hypothesis formulation and experimentation to production deployment.
Currently, I contribute to several innovation projects, including EmotionsAI by AB Tasty, a real-time visitor segmentation solution based on emotional needs, designed to drive more effective content delivery.
End-to-End AI & Research Projects
Across these projects, I design and ship AI features that transform complex signals—behavioral, visual, or semantic—into usable product insights. The emphasis is on fast iteration, interpretability, and real-world deployment.
All projects presented here were fully designed and implemented independently, end to end.
My work is primarily grounded in Python, Google Cloud Platform, and modern ML libraries such as scikit-learn and TensorFlow.
Flagship project:
Predict visual attention areas from a single image to optimize designs before production.
• CNN-based visual attention prediction
• From state-of-the-art scientific models → interactive demo
Tags: Visual Attention · Computer Vision · UX Prediction · Transformers
Stack: Python · TensorFlow · Vertex AI · FastAPI
Featured project:
Search and explore visual content using multimodal embeddings, combining computer vision, semantic search, and scalable cloud deployment.
-
Multimodal embeddings (text + image)
-
Vector search with scalable indexing
-
End-to-end pipeline: ingestion → embedding → retrieval
Tags: Multimodal Search · Computer Vision · Semantic Search · Retrieval
Stack: Python · TensorFlow · Vector DB · Google Cloud Run · FastAPI
Featured project:
Predict user engagement in the first seconds of a session using mouse movement patterns, enabling early and actionable personalization.
-
Behavioral signals extracted from real user interactions
-
Feature engineering on mouse trajectories
-
From research prototype → real-time inference API
Tags: Behavioral ML · User Interaction Modeling · Engagement Prediction
Stack: Python · Scikit-learn · XGBoost · Google Cloud Run · FastAPI
Automatically evaluate UX dimensions from website screenshots using deep learning, turning qualitative design principles into quantitative signals.
-
Trained on curated datasets of real web interfaces
-
Image-based feature extraction with CNN architectures
-
From UX research concepts → product-oriented demo
Tags: UX Prediction · Behavioral ML · Computer Vision
Stack: Python · TensorFlow · Scikit-learn · Vertex AI · FastAPI
👉
Product & Delivery
Product-Oriented ML
-
Turned complex signals (visual, behavioral, semantic) into actionable product insights
-
Worked closely with product & innovation teams
-
Focus on real-world constraints: latency, interpretability, usability
Project Management & Roadmapping
-
Defined research roadmaps
-
Prioritized initiatives based on impact, feasibility, and constraints
-
Broke down complex projects into actionable tasks and milestones
-
Coordinated with product, engineering, and research stakeholders
Project Ownership
-
Designed and implemented multiple projects end-to-end, solo
-
From research idea → deployed demo → product-ready system
⭐
👉
👉
👉
Core Tech Stack
Languages & Data
-
Python · R · SQL
Machine Learning
-
TensorFlow · Keras · PyTorch · Scikit-Learn
Data & Visualization
-
NumPy · Pandas · Matplotlib · Seaborn
APIs & Backend
-
FastAPI · REST APIs
Cloud & Infrastructure
-
Google Cloud Platform (Cloud Run, Vertex AI, GCS, BigQuery)
-
ClickHouse · Vector databases
MLOps & Deployment
-
Containerized deployment (Docker)
-
Model serving · Batch & real-time inference
-
Experiment tracking · Production monitoring
Education & Experience
Education
Experience
2025-Today
Research Engineer
R&D Department - AB Tasty
Built and shipped production-ready AI systems for high-traffic products serving millions of users.
-
5+ ML POCs delivered and 3 ML products shipped
-
Conducted extensive benchmarking across multiple models and pipelines
-
Explored and evaluated dozens of product-level features
-
Worked across core e-commerce use cases: segmentation, search and recommendation
Tech stack: Python · Machine Learning · FastAPI · GCP · Docker
2021-2024
Research Engineer (Phd Program)
Dotaki (Acquired by AB Tasty)
Built behavior-driven personalization features at the intersection of applied research, machine learning, and product.
-
Contributed to EmotionsAI, a patented real-time visitor segmentation system based on emotional needs.
-
Designed and engineered novel behavioral features integrated into production ML models
Tech stack: Python · Machine Learning
Tags: Behavioral Feature Engineering · Eye-Tracking · Inferential Statistics
2021-2024
PhD in Cognitive Sciences
Université Paris Cité - Vision Action Cognition Lab
Personality prediction based on eye movements and mouse behavior to personalize web page content.
Designed and led large-scale eye-tracking & mouse-tracking experiments (100M+ multimodal datapoints); co-authored a peer-reviewed scientific paper
(Scientific Reports, Nature, +4000 reads)
2019-2020
Cognitive Science Engineering
Master Degree
Institut Polytechnique de Grenoble
Engineering methods, neurosciences, machine learning, and artificial neural networks.
Improved model robustness by +46 accuracy points under adversarial attacks.
Tags: Python · Deep Learning · TensorFlow
2018-2019
Cognitive Psychology
Master Degree
Université Paris Descartes
Eye-tracking applied to user experience and online advertising.
Tags: Python · Inferential Statistics · Behavioral Science
SKILLS
From Code to Strategy: Tech, Management & Product Vision
⭐
Applied AI & Machine Learning
Applied Behavioral ML
-
Built end‑to‑end ML algorithms for UX prediction and personalization
-
Deployed real-time inference APIs under latency constraints
-
Developed ML models to predict visual attention using eye‑ and mouse‑tracking data
Computer Vision
-
Trained CNN-based models for ux prediction from screenshots
-
Worked with eye-tracking datasets (millions of datapoints)
👉
👉
Embeddings & NLP
👉
• Built representation learning pipelines (text, image, multimodal)
• Implemented vector search & similarity retrieval
• Designed semantic search & re-ranking systems
• Worked with CLIP & multimodal text–image models
Modeling & Experimentation
Behavioral Modeling & Feature Engineering
-
Designed features from mouse trajectories, temporal dynamics, interaction patterns
-
Built predictive models for early-session engagement
Research & Experimentation
-
Hypothesis-driven approach: problem framing → offline validation → production
-
Translated research prototypes into interactive demos & APIs
-
Strong focus on interpretability and robustness
UX Research & Experimental Design
-
Administered questionnaires (psychometrics, UX metrics, behavioral scales)
-
Defined experimental protocols (hypotheses, variables, controls, validation criteria)
-
Bridged UX research insights → ML features → product decisions
⭐
👉
👉
👉
Production, Cloud & Deployment
Cloud & Infrastructure
-
Model serving & API design for high-traffic environments
⭐
