Yanis Amirou

Solar Energy Production Forecasting Context: Industrializing R&D forecasting algorithms into a robust MVP to optimize energy management for photovoltaic plants. Key Achievements: Leadership: Defined the MVP roadmap (Jira) and mentored a team of junior Data Scientists. Modeling: Trained and optimized (Bayesian Optimization via SageMaker) boosting models (XGBoost, LightGBM) and Deep Learning models for complex time series. Validation of results. End-to-End MLOps: Built a complete CI/CD pipeline Github Actions, Docker images, Airflow orchestration. Implemented Observability for data drift and concept drift detection. Impact: Successful production release of the automated forecasting pipeline with full quality monitoring (Drift detection). Stack: Python, AWS (SageMaker, EKS), Github actions, Airflow, MLflow, Docker, XGBoost/LightGBM.

Smart Wearables & Biomechanics Context: Zhortech designs smart insoles capable of analyzing walking and running patterns to help detect certain diseases and to assess or improve athletes’ performance. The challenge was to transform raw sensor data into reliable biomechanical indicators. This required overcoming the physical limitations of the sensors (noise) through Deep Learning. Signal Processing & DL: Developed Transformer models for IMU signal denoising and pathology detection (XGBoost), significantly improving ML performance across all solutions. GenAI Assistant (POC) : Built an internal RAG chatbot (Llama 2, LangChain, ChromaDB, streamlit) to query technical documentation. Data Augmentation: Used GANs and Reinforcement Learning to generate synthetic data for rare clinical conditions. Impact: Deployed clinical KPI dashboards (Streamlit/Power BI) and mentored junior staff. Stack: Python, TensorFlow, PyTorch, Llama 2, LangChain, ChromaDB, AWS, GANs.

Spatio-Temporal Forecasting Context: Forecasting on interconnected sensor networks by exploiting underlying graph structures. Key Achievements: Hybrid SOTA Architecture: Designed models combining Graph Convolutional Networks (GCN) for spatial dependencies and Transformers for temporal dynamics. Use of Topological Data Analysis (TDA) for robust structural feature extraction. Impact: Achieved State-of-the-Art performance on graph time series forecasting, validating a new modeling approach for energy grids. Stack: Python, TensorFlow, PySpark, GNN, Transformers, TDA, Scikit-learn.

Research in machine learning, time series, signal processing and topological data analysis - Publication: S-Transform with a Compact Support Kernel and Classification Models Based Power Quality Recognition. - Teaching: Linear algebra and calculus

Forecasting Paris air pollution time series - Statistical analysis of weather and traffic time series data - Feature engineering - Prediction of NO2 rate in Paris air - Trained different models like random forest, svm, NeuralNets

Teaching: - Linear Algebra - Multivariate analysis and integration - Group theory - Calculus