Freelancer profile translated to English.

Description

Lead Tech Data & AI - ML/DL Expert:Pragmatic and visionary with 10+ years bridging cutting-edge research and production ML/AI systems. Started from computational biology, developing innovative algorithms, I evolved into leadership that not only understands ML/AI theory—I've built teams that deploy models at scale. From architecting MLOps platforms serving millions of daily predictions to mentoring data scientists into ML engineers, I believe great AI engineering combines experimental rigor with production reliability. Currently seeking opportunities to lead data/AI projects and teams where scientific rigor meets business impact and innovation drives measurable results.

Industry field of expertise

Languages

French
Native or bilingual
English
Fluent

Workplace preferences

Can work on-site

Lyon (up to 50km)

Connect Technology
Chief Technology Officer
RETAIL (SMALL BUSINESS)
January 2023 - December 2025 (2 years and 11 months)
Lyon, France
Leading AI/ML strategy and building intelligent systems for the transformation of gaming & retail in emerging markets.

ML Intelligence Platform: Architected an end-to-end ML platform processing 1M+ daily transactions with real-time fraud detection, demand forecasting, and customer behavior models. Built a data mesh architecture enabling 6 ML models in production with 99.8% uptime and sub-100 ms inference latency.
Data Science Team Building: Built and led 2 ML engineering squads (12 engineers: 5 data scientists, 7 ML engineers), establishing best practices for experiment tracking, model versioning, and A/B testing. Created a career path between data science and engineering, reducing time-to-production from months to weeks.
Real-Time ML Pipeline: Designed a streaming ML architecture with Spark Structured Streaming for real-time feature calculation and model inference. Reduced prediction latency from 4 hours to 5 minutes, enabling dynamic pricing and inventory optimization, increasing revenue by 15%.
ML Quality Framework: Implemented comprehensive ML observability with model performance monitoring, data drift detection, and automated retraining pipelines.
Cloud ML Infrastructure: Led migration to a cloud-native AWS ML platform with SageMaker, EMR, and S3, using infrastructure-as-code via Terraform. Achieved a 30% cost reduction through spot instances and autoscaling while maintaining 95% model availability.
Responsible AI Practices: Established a model governance framework with lineage tracking, bias monitoring, and explainability requirements. Ensured regulatory compliance while making ML models transparent and interpretable for business stakeholders and auditors.
Technologies:

. Python/Scikit-learn
• MLflow/Weights&Biases
• A/B Testing
• Spark MLlib
• Feature Stores
• Team Leadership
• AWS SageMaker
• Model Monitoring
• ML Strategy
Machine learning Project Management Team Management Artificial Intelligence Scala
Skillscaper
Co-Founder & Chief Technology Officer Skillscaper
EDUCATION AND E-LEARNING
November 2023 - May 2024 (6 months)
Lyon, France
Building an AI-native skills assessment platform combining symbolic AI and modern LLMs.

Hybrid AI Architecture: Designed an innovative assessment system combining neural (LLMs) and symbolic (Prolog) approaches for context-agnostic skill evaluation. Processed 500K+ dialogue records with Spark NLP for feature extraction and custom inference engines for logical reasoning, achieving 92% accuracy vs. 78% baseline.
Production ML Pipelines: Built end-to-end Scala/Spark feature engineering pipelines with feature store integration, ensuring training-serving consistency. Implemented automated feature validation and monitoring, improving model accuracy by 25% and reducing debugging time by 60%.
MLOps: Established a complete MLOps stack including experiment tracking (MLflow), model registry, CI/CD for models, and automated testing. Reduced model deployment cycle from 2 weeks to 2 days while maintaining rigorous validation and rollback capabilities.
LLM Engineering: Integrated and fine-tuned LLaMA models for dialogue generation and evaluation, implementing prompt engineering and response validation frameworks. Combined with Claude AI for reasoning tasks, creating a hybrid system balancing flexibility and reliability, achieving a 40% cost reduction compared to a pure LLM approach.
Scalable NLP: Optimized Spark NLP pipelines for large-scale text processing, implementing efficient tokenization, embedding generation, and entity extraction. Reduced processing time by 40% through strategic caching, broadcast joins, and partition optimization while handling growing volumes.
Real-Time ML Dashboard: Developed a real-time analytics platform tracking model performance, feature distributions, and business metrics. Enabled immediate feedback loops for continuous model improvement and A/B testing of assessment strategies.
Moobifun
Chief Technology Officer
SOFTWARE PUBLISHING
January 2020 - April 2023 (3 years and 3 months)
Lyon, France
Leading Tech/ML/AI strategy for large-scale personalized gaming experiences.

ML Platform Modernization: Led the transformation from rule-based to ML-driven personalization, migrating to a Spark ML platform with Delta Lake for feature storage. Scaled from 1TB to 10TB daily, supporting 15+ production models serving 10M+ predictions/day with 99.5% availability.
Tech Team Growth: Grew the Tech team from 20 to 35 practitioners (12+ Dev, 5 DevOps, 10 ML engineers+ data scientists), establishing research-production separation while maintaining collaboration. Implemented code reviews, model reviews, and knowledge sharing, improving model quality by 50%.
Personalization Engine: Designed and deployed a recommendation system using collaborative filtering, deep learning embeddings, and contextual bandits. Models powered dynamic game recommendations, increasing engagement by 40%, retention by 25%, and revenue by 18% through personalized experiences.
Model Performance Optimization: Systematically optimized inference pipelines, achieving a 35% cost reduction through model compression, batching strategies, and efficient feature computation—while maintaining 99.5% SLAs and improving prediction latency from 200ms to 50ms.
Feature Store Architecture: Implemented a production AWS S3 feature store with Apache Hudi, enabling feature reuse across teams, point-in-time correctness, and low-latency serving.
ML Observability: Built comprehensive ML system monitoring tracking data quality, model performance, feature distributions, and business impact. Reduced average detection time by 80% and average resolution time by 65% for ML incidents through automated alerting and root cause analysis.
Technologies:

• Recommendation Systems
• ML Monitoring
• Spark MLlib
• Feature Stores
• Team Scaling
• Deep Learning
• Model Compression
• ML Production