Eléonor Kioulou

I completed my six-month final internship at Mandala AI, a Thai company specializing in artificial intelligence and data analysis. My role as a Data Scientist involved developing predictive models on French-language text data. These models were designed to be integrated into the company’s data analysis applications in the future, particularly for social media data. The goal was to enhance their ability to analyze and interpret trends and sentiments from social media platforms in a more efficient and accurate way. What I did : Developed a sentiment analysis model using CamemBERT to classify French social media comments into positive, neutral, and negative sentiments. Achieved an 88% F1-score after fine-tuning and addressing data imbalance through oversampling techniques. Optimized model performance by fine-tuning hyperparameters (batch size, learning rate) and integrating ensemble methods (FlauBERT, XLM-Roberta), reducing overfitting by 22% and improving cross-platform generalization. Deployed a Streamlit application to compare predictions across multiple AI models, cutting evaluation time by 50% and enabling real-time testing for end-users. Processed and analyzed 3,400+ French comments, automating data preprocessing workflows (text normalization, label encoding) to accelerate pipeline speed by 40%. Manually labeled and balanced the dataset to ensure high-quality training data, addressing challenges like sarcasm and platform-specific language variations.

I started as an intern in 2021 and then continued as a temporary employee during certain periods.

My mission involves the development and promotion of an innovative video conferencing solution. I am responsible for prospecting interested businesses, developing advanced features in Python, such as replacing the background with an image or GIF, and creating real-time subtitles for video calls. My goal is to enhance remote communication by making it more efficient, enjoyable, and accessible to everyone.