🎓 MSc Applied Mathematics & Statistics — Data Science
Université de Caen Normandie · France

Building reliable Data & AI systems — end to end
ML pipelines · Multi-agent AI · Cloud deployment · Production monitoring

Clean pipelines · Reproducible results · Real business impact

• 🧠 Applied AI & LLM systems — multi-agent orchestration, RAG, guarded Text-to-SQL and conversational memory
• 🤖 Machine Learning & NLP — classification, fraud detection, imbalanced learning, technical-log analysis and clustering
• ⚙️ ML Engineering / MLOps — APIs, containerization, cloud deployment, experiment tracking, monitoring and alerting
• 🔌 Data Engineering — ETL/ELT pipelines, BigQuery, SQL, REST APIs and scheduled cloud automation
• 📈 Time series — forecasting, feature engineering, leakage-free evaluation and model comparison
• ☁️ Cloud — GCP for Data & AI systems; Azure ML, Azure Databricks and Azure AI Foundry

Predictive ML  → data → features → training → evaluation → deployment → monitoring

Applied AI     → business question → agent routing
               → RAG / guarded Text-to-SQL / integrity controls → sourced answer

Data systems   → APIs & files → validation → BigQuery → scheduling → alerts

Not just notebooks — reliable systems designed around real business needs.

March – August 2026 · Mondeville, France

Built Data & AI solutions for accounting and Order-to-Cash processes, combining production cloud automation with internally deployed AI and machine-learning pilots.

• REFLEX: designed a secure, read-only multi-agent assistant combining guarded Text-to-SQL, RAG, deterministic integrity checks, conversational memory and Cloud Run deployment.
• SAP rejection diagnostics: built an NLP classification and diagnostic pipeline combining TF-IDF, logistic regression and business controls — 89.5% cross-validated accuracy and 86.1% macro-F1.
• Cloud data automation: deployed scheduled Gmail and Cloud Storage to BigQuery workflows with preprocessing, freshness checks, controlled reprocessing and Google Chat alerts.
• Business controls: automated VAT verification through Cyplom/VIES, Trustpair alert routing and resilient legal-data extraction through the INPI API.

The internal source code and business data remain confidential. A sanitized portfolio presents the architectures, responsibilities and validated results:

➡️ Explore the Data & AI internship case studies

Tech: PySpark · Spark MLlib · Python · SQL · Superset

• Distributed pipeline: join, clean and prepare identity + transaction data at scale
• Feature engineering, supervised modeling, imbalanced learning
• Evaluation: ROC-AUC, PR-AUC, recall, F1 + dashboard for fraud pattern analysis

➡️ github.com/elfahad98/ieee-fraud-pyspark

Tech: Python · scikit-learn · XGBoost · MLP · pandas · NumPy

• Behavioral and time-based feature engineering (frequency, device signals, login patterns)
• Benchmarking: Logistic Regression vs XGBoost vs MLP
• Imbalanced learning, threshold tuning, recall-first optimization

➡️ github.com/elfahad98/ato-fraud-detection-mlp

Tech: Python · pandas · scikit-learn · statsmodels · MLflow

• Preprocessing pipeline, lag and rolling-window features, leakage-free splits
• Model comparison: Ridge vs SARIMA/SARIMAX vs baselines
• Experiment tracking with MLflow, evaluation on validation and test sets

➡️ github.com/elfahad98/water-temp-forecast

• 📧 Email: elfahad98@gmail.com
• 🐙 GitHub: github.com/elfahad98