AF Consult

Apollon AI

The ApollonAI project is a framework designed to simplify the development of AI models while promoting best practices and ensuring seamless integration with cloud applications. It provides a well-structured project architecture to facilitate scalability and maintainability. Built with Python, Rust, and C++, the framework offers flexibility and performance, making it suitable for diverse AI applications. ApollonAI focuses on streamlining the model-building process while maintaining high standards for code quality and cloud connectivity, empowering developers to create efficient, production-ready AI systems.

Autonomous Vehicule Image Processing

The project focuses on building an image processing pipeline for an autonomous vehicle system, leveraging state-of-the-art segmentation models like U-Net, SegNet, and DeepLabV3+. The goal is to preprocess data, develop and optimize a deep learning model for segmentation, deploy a prediction API using Flask or FastAPI on a cloud platform, and integrate it with a web application for real-time image predictions. The project also includes full documentation, model analysis, and presentation of the results for review.

May 5, 2025 0

AI-based Fashion Recommendation System

Fashion-Insta, a company in the fashion industry, aims to launch a mobile application that leverages artificial intelligence to provide users with fashion recommendations. The app allows users to take photos of their favorite outfits and receive suggestions for similar clothing items. As an AI Product Manager, you are responsible for leading this project, working under the guidance of Alicia, the VP of Product. The app will utilize cloud technologies from Microsoft Azure, and the project follows an agile SCRUM methodology. The project scope includes defining user stories, prioritizing them, estimating costs, and ensuring GDPR compliance, particularly regarding personal data collected from user images. A key objective is to demonstrate the app’s profitability and technical feasibility to the executive committee (COMEX) within three weeks. Your responsibilities include presenting a detailed project plan, outlining potential risks, and providing mitigation strategies to gain approval for this innovative AI-driven fashion application.

Home Credit Scoring Model

Credit scoring model for a financial company providing consumer loans to clients with limited credit history. The goal was to create an interpretable model to predict loan repayment likelihood and support loan approval decisions. Using historical loan data, financial details, and borrower behavior, I engineered new features and addressed data imbalance with techniques like SMOTE. Models such as Logistic Regression, Decision Trees, and Gradient Boosting were trained, with cost-sensitive learning applied to optimize the balance between false negatives and false positives. The final model achieved an AUC of 0.81, and SHAP values were used to ensure model interpretability, aiding customer relationship managers in making informed loan decisions. The project involved skills in data and predictive modeling, Scikit-learn, XGBoost, Random Forest, Docker, and Python.

Sentiment Analysis Using Deep Learning

The project involves developing a sentiment prediction AI model for "Air Paradis" to anticipate negative buzz on social networks. The task includes creating a prototype API to predict tweet sentiment using both simple and advanced models, including deep learning and BERT-based approaches. Additionally, the project will implement MLOps best practices, including experiment tracking with MLFlow, continuous deployment, and production monitoring with alert systems. The final deliverables include a cloud-deployed API, scripts for the models, an API test interface, and a blog article documenting the process.

Participate at Smart City Competition

The project involves participating in a data science challenge organized by the city of Paris, in collaboration with the NGO -Data is for Good.- The task is to conduct an exploratory data analysis on a dataset of trees in Paris, as part of the -Végétalisons la ville- program. The goal is to optimize tree maintenance routes, reducing travel time and improving the care of the city's trees. The deliverables include a presentation summarizing the dataset and methodology, and a Jupyter Notebook with detailed documentation of the analysis steps for a non-technical audience.

Code Stream

CodeStream is a comprehensive project management software that enforces a specific architecture, coding conventions, and a unique ticket management system. It offers advanced features for project tracking, development workflows, and centralized control, ensuring consistency, efficiency, and scalability across projects.

Scalable Big Data Processing for Intelligent Fruit Classification

The project involves developing a Big Data processing pipeline for "Fruits!", a startup in the AgriTech sector focused on creating intelligent fruit-picking robots. The initial goal is to launch a mobile application that allows users to photograph fruit and receive detailed information, helping to raise awareness about fruit biodiversity. This app will also serve as a foundational step in building a fruit image classification engine. The project leverages a cloud-based Big Data environment using AWS EMR and PySpark to process fruit images and labels, ensuring scalability for future data volume growth. The focus is on creating an efficient data processing architecture, ensuring GDPR compliance, and optimizing cloud costs.

Projects

Apollon AI

Project Mental Model Manager