Med Abdelkader KETATA

Development of a chatbot for project scaffold generation: Designed and implemented an intelligent conversational assistant that automates the creation of project scaffolds, enabling the generation and modification of APIs (OpenAPI specifications in JSON and YAML) from natural language requests. Integrated into an existing Spring Boot application, the project involved building, cleaning, and structuring datasets, as well as performing fine-tuning, optimization, and benchmarking of LLMs (StarCoder, TinyLlama) using Hugging Face with advanced techniques such as PEFT and LoRA to improve efficiency and memory usage. Implemented an NLP multi-model pipeline with Flask and vLLM, developed the Angular frontend, enhanced the Spring Boot backend, and configured Keycloak along with collaborative project management tools. Finally, deployed the full solution with Docker and Docker Compose to ensure portability and reproducibility.

I participated in the development of a dynamic form mobile application, using Ionic for the frontend and Spring Boot for the backend, with a PostgreSQL database. The application consists of two parts: the user section and the admin section. In the admin section, the administrator manages the forms, the fields within each form, and their types, while also reviewing user responses. As for the user section, it allows users to view the forms created by the administrator, select a specific form, and submit their responses.

I contributed to the development of a chatbot that assists platform users on Rec Inov and helps candidates throughout the recruitment process by matching them with suitable positions. The frontend was built with ReactJS, the backend with Node.js, and MongoDB was used as the database. The machine learning component was implemented using the Rasa framework.

I contributed to the improvement of a CV parser web application by working on three aspects: Parsing, Classification, and Similarity. My tasks included data preprocessing, setting up and training a multi-class classification AI model, and enhancing the frontend using the Streamlit framework.