Freelancer profile translated to English.

Description

Business Impact & Production-Oriented Data Scientist – Explainable AI, MLOps & Cloud

Data Scientist with **2 years of experience in an industrial environment**, I support companies in the design, industrialization, and deployment of reliable, explainable, and directly usable AI solutions for business teams.

AtAlstom**, I contributed to **predictive maintenance (CBM)projects that led to:

📈+15% defect detection

⏱️ -10 hours of diagnosis per incident

🤖 Robust and explainable models, integrated into real operational processes

I combine solid technical expertise(Python, ML/DL, LightGBM, Spark, LLM, RAG)with a product and business vision, essential for transforming a functional need into a deployed solution.

💪 Strengths

- Production-oriented AI, not just notebooks

- Native MLOps & Cloud: MLflow, Docker, Flask APIs, GCP / AWS

- Explainable & traceable AI, adapted to industrial and regulated contexts

Ability to simplify complex concepts and collaborate with non-technical teams (finance, industry, public services)

🔴 Use Cases Covered

- Predictive maintenance & anomaly detection,

- Credit scoring & decision support,

- Decision-making dashboards & data storytelling,

- Intelligent chatbots (LLM, RAG, LangChain, FAISS),

- Big Data & distributed computing pipelines (Spark, EMR)

🎯 Objective:

Deliver useful, understood, and adopted data solutions with a real return on investment.

Languages

French
Native or bilingual
English
Fluent
Arabic
Basic

Workplace preferences

Can work on-site

Lyon (up to 10km)

Alstom Group Crespin
Data Scientist
MECHANICAL ENGINEERING
September 2023 - September 2025 (2 years)
Valenciennes, France
TGV2020 Project – DATA ANALYSIS: Condition-Based Maintenance (CBM)

✅ Analysis of toilet data from a TGV's digital calculator, enabling the detection of tank leaks (WWT, GWT), curve issues, etc., using Condition-Based Maintenance (CBM).

✅ Classification of toilet pressure defects (multimodal model)

✅ Analysis of gray water tightness (GWT sensors) Optimization of diagnostic times: -10 hours per incident.
OpenClassrooms
Implement a scoring model
BANKING AND INSURANCE
October 2024 - December 2025 (1 year and 2 months)
Valenciennes, France
✅ Build a scoring model that automatically predicts the probability of a customer's bankruptcy.

✅ Analyze the features that contribute most to the model, both generally (global feature importance) and at the customer level (local feature importance), to allow a case manager to better understand the score assigned by the model, ensuring transparency.

✅ Deploy the prediction scoring model using an API and create an interface to test this API.

✅ Implement an end-to-end MLOps approach, from experiment tracking to production data drift analysis.
MLOps Airflow flask Google Cloud Platform (GCP) Streamlit
OpenClassrooms
Segment e-commerce customers
E-COMMERCE
January 2025 - March 2025 (2 months)
Lyon, France
✅ Define the strategy for developing an unsupervised learning model, select and train models suitable for a business problem to perform segmentation or data reduction.

✅ Evaluate the performance of unsupervised learning models based on different criteria (scores, training time, etc.) by adjusting parameters to choose the most performant model for the business problem.
Customer segmentation Clustering Machine learning Deep Learning SQL