My Projects
A showcase of my academic and personal projects in computer science, machine learning, and web development
Soccer - Computer Vision and Player Tracking
I am developing an advanced computer vision system for soccer analysis that detects players on the field, tracks their movements over time, and evaluates the danger level of different in-game actions. Using YOLO-based detection, multi-object tracking, and field-mapping techniques, the system follows players, identifies dangerous zones, and analyzes how actions evolve in real time. This includes recognizing when players enter high-threat spaces, when passing opportunities appear, and how team shape changes during an attack. Beyond sports analytics, this project contributes directly to the AI field by pushing the limits of real-time detection, spatial reasoning, and context-aware decision modeling. It serves as a practical testbed for combining object detection, tracking, and predictive modeling into a single intelligent pipeline. The techniques developed here—such as dynamic threat estimation, trajectory-based action scoring, and multi-agent interaction analysis—can be reused in autonomous systems, robotics, and video understanding research. This ongoing work helps expand how AI interprets complex human movement and tactical patterns in dynamic environments.
Revenue Prediction - Orange Money
I developed a complete predictive analytics system to forecast the monthly Revenue of multiple telecom services at Orange Côte d’Ivoire. The project involved building a scalable forecasting pipeline using Facebook Prophet, enriched with domain-specific regressors, seasonal patterns, national holidays, and historical growth signals. I engineered a custom module, PredictionCAUtils, responsible for preparing features, generating future datasets, and producing accurate prediction ranges for services such as CANAL and E-Commerce. The system was designed to integrate seamlessly with operational workflows and provide actionable decision-support for commercial strategy.
Inspection Drone – Thermal Overheating Detection
In this project, I developed an intelligent inspection system that uses thermal imagery captured by drones to detect overheating in critical telecom infrastructure components. The goal was to automatically identify abnormal temperature levels on pylons and equipment by combining computer vision with thermal analysis. The system classifies temperature ranges, estimates the global temperature of the scene, and evaluates each detected object relative to its environment to identify possible failures. I trained a YOLOv8 detection model on a thermal dataset to recognize key objects on drone-captured images and to extract their corresponding temperature readings. Using a custom thermal-color segmentation technique, I converted color gradients into temperature intervals to estimate the global temperature of each scene. I then implemented a deviation-based diagnostic rule
Genora – RAG-Powered AI Assistant for Human Resources
Genora is an intelligent HR assistant that I designed using a Retrieval-Augmented Generation (RAG) architecture to streamline internal communication and support employees within an organization. The system integrates natural language understanding with a dynamic document-retrieval pipeline, allowing Genora to answer questions accurately based on official HR policies, internal guidelines, onboarding documents, and company procedures. Its goal is to provide instant, consistent, and reliable support for day-to-day HR requests while reducing workload for human HR teams. By combining a vector database, document chunking, and an LLM-driven reasoning engine, Genora can interpret queries, retrieve the most relevant information, and generate precise, context-aware responses. The assistant was designed for tasks such as explaining benefits, guiding employees through administrative steps, answering leave-related questions, and providing quick access to key HR documents. Genora improves accessibility, reduces response time, and enhances the overall employee experience through an AI solution that is both flexible and secure.
Face Mask Detection through Image classification
This project was created in response to the global health crisis that began in 2019, a pandemic that revealed how unprepared societies were for large-scale viral outbreaks. Although we are no longer in that situation, the goal of this work is to help the world be better equipped for future public health emergencies. Using deep learning, the project focuses on building a system capable of identifying individuals who are not wearing a face mask or who are wearing it incorrectly. By automatically detecting improper mask usage in public spaces, the system can support health officials and organizations in reducing transmission risks and enforcing safety measures more efficiently. The solution is powered by convolutional neural networks (CNNs) built with TensorFlow and Keras, combined with OpenCV for real-time image processing. The model classifies images into multiple categories—such as “mask,” “no mask,” or “incorrect mask”—and can be deployed in cameras, kiosks, or monitoring systems. Through Python-based data preprocessing, augmentation, and model training, the system achieves strong accuracy and robustness across varied lighting conditions, angles, and environments. This project demonstrates how deep learning can play a vital role in strengthening public health readiness and improving early-response strategies during future outbreaks.
Java Desktop Application
This project focuses on the large-scale visualization and exploration of a countrywide car accident dataset covering 49 U.S. states. The dataset contains 1.5 million accident records collected between February 2016 and December 2020 through a network of traffic-related APIs. These APIs aggregate real-time incident reports from diverse sources, including state and federal departments of transportation, law enforcement agencies, traffic cameras, road sensors, and other monitoring systems. By consolidating these multi-source streaming feeds, the dataset provides a comprehensive view of accident patterns, environmental conditions, temporal trends, and geographic distributions across the country. Using Tableau, Power BI, and Exploratory Data Analysis (EDA) techniques, the project transforms this massive dataset into interactive dashboards and visual narratives. These visualizations allow users to uncover key insights such as high-risk regions, peak accident hours, seasonal trends, weather-related incidents, and correlations between traffic density and accident severity. The resulting dashboards help stakeholders—such as public safety agencies, transportation planners, and policymakers—better understand the factors that influence road safety and make data-driven decisions to improve traffic management and reduce accidents nationwide.