Younes Amoussas

I designed a mobile app for 3D virtual try-on using photogrammetry and augmented reality to enhance the online retail experience, focusing on architecture and specifications for scalability. Achieved Tasks: - Designed frontend (React Native with ARKit/ARCore, Three.js) and backend microservices (Python/Node.js) architecture. - Drafted specifications for 3D avatars (MediaPipe, TensorFlow), including latency and scalability constraints. - Proposed AWS cloud infrastructure (ECS/EKS, Lambda, Terraform) for deployment. - Collaborated with UX/UI and business teams to align technical and operational goals.

I developed an automated system to classify mobile users' web browsing history on Google Cloud Platform (GCP), replacing costly semi-manual scripts, with the goal of optimizing activity categorization through web scraping, LLMs, and deep learning while cutting expenses. Achieved Tasks: - Performed web scraping to extract hostname metadata (content, categories) using Python (BeautifulSoup, Scrapy) for data enrichment. - Designed an initial scoring system, followed by ML models (Random Forest, XGBoost) and deep learning (CNN, Transformers) for final categorization, leveraging LLMs (Hugging Face) for semantic analysis. - Implemented Spark pipelines on GCP (Dataflow, BigQuery) to handle massive data volumes. - Optimized automation to reduce execution costs by 12 % ($10K/month).

I built and deployed a B2B recommendation system based on collaborative filtering, leveraging user, product, and purchase databases to improve conversion rates through personalized and scalable recommendations. Achieved Tasks: - Developed a collaborative filtering model to suggest products based on user preferences and purchase history. - Performed data integration and enrichment via APIs and databases to create a 360° customer view. - Deployed the system in production on AWS (SageMaker, Lambda) with MLOps pipelines for scalability and low latency. - Conducted A/B tests to validate models and created dashboards to monitor performance metrics (accuracy, conversion).

I designed an end-to-end system to analyze and predict US stock performance for 1000 stocks, incorporating historical data (via APIs), watchlists (Excel/PDF), and algorithmic trading strategies. The goal was to generate annual predictions, buy/sell signals, and insights through ML models. Achieved Tasks: - Developed models (XGBoost, ARIMA, deep learning) for predictions and buy/sell signals, including feature engineering (200+ variables), hyperparameter tuning, and cross-validation. - Performed data analysis with enrichment via web scraping and integration of time-series data, consolidating watchlists (Excel/PDF). - Conducted backtesting of algorithmic trading strategies. - Deployed the system in production on cloud infrastructure, with weekly/monthly execution and dashboards for monitoring.

I analyzed data from 1.2 million client debt transactions to predict payment defaults and optimize the collection process, aiming to increase recovered funds through a predictive analytical approach. Achieved Tasks: - Performed data analysis and exploration (EDA) of 1.2M transactions via BigQuery, extracting behavioral insights using KMeans clustering. - Developed a classification model to predict default probabilities. - Optimized the collection process and created Looker Studio dashboards for business reporting and performance monitoring.

I enhanced and optimized an ML/AI and operations research-based software for managing European deliveries from warehouse to last mile, implementing new features and optimization methods drawn from the latest research papers. Achieved Tasks: - Conducted deep dive analysis on Big data (Redshift, SQL) to assess data quality and veracity, implementing custom metrics. - Optimized the AI decision-making and delivery management tool, reducing runtime through new feature deployments. - Formulated and backtested 20+ business scenarios using the optimized tool to validate the strategy. - Visualized and presented results via QuickSight dashboards and detailed reports for stakeholders.

• Set up a data collect and analysis solution based on condition monitoring and machine learning algorithms to predict machines' potential failures • Communicate and work effectively with different parts and teams involved in the project on a weekly basis • Synthesize clear conclusions from complex information in order to simplify the solution's concept for other departments

• Define and analyse the client's need • Develop a website and an intranet