Oussema Kirmani

– Developed deep learning models for human activity recognition using two distinct datasets. – Designed and implemented a CNN-LSTM architecture for activity classification. – Achieved 93.67% accuracy in predicting six activity classes. – Conducted data preprocessing, cleaning, and feature engineering to enhance model performance. – Developed a CNN×LSTM architecture to recognize human actions from skeleton data. – Reached 79.45% accuracy after optimizing the model and preprocessing steps. – Performed model evaluation, architecture optimization, and accuracy assessment for both datasets. – Applied end-to-end deep learning pipeline development from raw data to final performance metrics.

– Developed a domain-specific Medical NER system to improve the semantic understanding of a chatbot and optimize the structure and accuracy of its underlying knowledge graph. – Designed and implemented two specialized NER models. – Gene/Protein Entity Model: Achieved an F1-score of 78.8 on benchmark datasets. – Disease/Chemical Entity Model: Achieved an F1-score of 82.4, outperforming several baseline models. – Improved entity linking and chatbot response generation via structured integration with a knowledge graph. – The project was selected by a jury for proposed publication in IEEE, recognizing its research contribution and implementation quality. – This Project got the best grade in the promotion, 17.5/20

– Designed and developed a module for a web application, focusing on creating an efficient and scalable component to enhance functionality and user experience. – Implemented automation processes to streamline workflows, optimize performance, and ensure seamless integration with the existing system. – This project deepened my expertise in web development, automation tools, and collaborative software engineering practices.

– Developed a web application that manages intern data, using React for the frontend and Node.js for the backend.