Roxane Kouamé

Creation, annotation, and labeling of multimodal datasets (text, image, audio, video) for the training and benchmarking of AI models: LLM, VLM, OCR, ASR, Speech-to-Text. Participation in the RLHF (Reinforcement Learning with Human Feedback) process through comparative evaluation of model-generated outputs (preference, relevance, coherence, ethics). Performance analysis of generative and conversational models (GPT-4, Claude 3, Gemini 1.5, Llama 3, Mistral, Whisper) based on accuracy, consistency, and contextual robustness criteria.

Contributed to the redesign of the user interface for a hospital stock management module using ASP.NET Web Forms and Bootstrap 5 to improve ergonomics, accessibility, and multi-platform compatibility. Migrated existing elements to Bootstrap 5 components and implemented flexible grids, responsive forms, and data tables with filters and pagination. Adapted the structure to ASP.NET Master Pages with server-side event handling. Added interactive features using JavaScript and jQuery (modals, alerts, dynamic notifications). Collaborated with the technical team via Git/GitLab/TortoiseSVN, participated in code reviews, and integrated changes within an Agile Scrum cycle. Authored technical documentation and user guides to facilitate future maintenance. Tools used: ASP.NET Web Forms, C#, HTML5, CSS, Bootstrap 5, JavaScript, jQuery, Git/GitLab, Visual Studio 2022, Figma, Agile Scrum methodology.

Designed and developed a complete image-processing pipeline for SEM images to automate diamond wafer quality inspection: acquisition, preprocessing, segmentation, and classification. Implemented and compared multiple approaches: K-Means, DBSCAN (introduced through a personal tech watch), and U-Net for fine-grained crystallographic defect segmentation. Extracted geometric features (area, circularity, mean intensity) using scikit-image and trained supervised classifiers (KNN, RandomForest) for automatic defect categorization. Built an interactive Python interface (Tkinter) to visualize, annotate, and export results as CSV reports and histograms. Optimized models and conducted multi-metric evaluations (Accuracy: 91.2%, F1-score, Recall), improving defect detection reliability and reproducibility. Tech stack: Python, OpenCV, scikit-image, scikit-learn, TensorFlow/Keras, Tkinter, NumPy, Pandas, Matplotlib.

During my internship at STMicroelectronics in Crolles, I worked on two key projects. The first project involved creating an interactive 3D model for the company's process academy. I used 3D modeling software such as SolidWorks to design semiconductor production equipment. I also integrated interactive elements into the model and conducted tests to validate its functionality. The second project focused on embedded programming, where I developed software for serial communication between a Micro Tool and a modern PC, using technologies such as Python, C/C++, Access Port, PyQt5, and openpyxl. These experiences allowed me to gain strong technical skills and work with cutting-edge technologies in the semiconductor industry.

Observation internship during which I acquired knowledge of 3D design and modeling software such as Autodesk Revit, AutoCAD, Robot, Inventor, and SolidWorks.