Yassine Handane

– Development of an educational content extraction pipeline inspired by MinerU architecture, including fine-tuning YOLO models for layout and border detection in educational documents. – Implementation of formula, table, and text detection modules adapted to educational content with post-processing systems to convert PDFs to structured formats (Markdown, JSON). – Optimization of multimodal recognition models for interpreting unstructured educational data via specialized Al agents (LLAMA, Mistral open source models and online services). – Development of automated test suites via prompt engineering for Al performance evaluation and ranking, enabling optimal model/prompt combination selection for educational content extraction. – Design and implementation of a RAG (Retrieval-Augmented Generation) system for automatic generation of personalized quiz questions for end users.

As part of the AISSPER project, I contributed to the development of an advanced Spoken Language Understanding (SLU) system for processing long-form content. Key achievements include: – Designed and implemented a sliding window mechanism to process sequences beyond the standard 512 token limit. – Adapted pre-trained models (DistilBERT for English, CamemBERT for French) for long-text processing. – Developed a model architecture integrating Masked Language Modeling (MLM) and Next Sentence Prediction (NSP) tasks. – Conducted comparative analysis of fine-tuning versus feature extraction approaches. – Enhanced ASR capabilities to handle extended speech inputs, improving transcription accuracy for long-form audio content. – Created a web application demonstrating the SLU system's capabilities.

I created user-friendly financial dashboards using Tableau, enabling effective data analysis and decision-making. I also documented and visualized complex data processes for the Data Factory, translating complex technical procedures into clear, understandable diagrams.

My end-of-year internship project is entitled « Mobile application for reporting about field incidents ». The main objective was to develop a mobile application that enables users to report incidents from the field. The application should have the functionality to allow users to create reports with details such as incident type, location, description, photos or videos, and other relevant information. Besides checking the progress of other reports.