Safouene Jebali

Benchmarking RAG system components (Vector databases, chunking algorithms, etc.). Implementing No-Code/Low-Code AI automation for AI agents, chatbots, RAG, and workflow automation. Technologies: Gradio, Zapier, OpenAI, LLMs, N8N

Design / implementation of Fine-Tuned Model for Multi-Modal Gen AI applied to ART Domain (Design, Interior Architecture, Outdoor Architecture) The purpose of the internship is the design and implementation of fine tuned models based in the training for base Gen AI multi-modal model on a private dataset of objects, the aim is to orient the model toward the generation of images based on the artist style and also embed the objects in various other context. Training and Inference Platform Preparation: Prepared and managed training and inference platforms using AWS and Runpod to support AI model development. Selected and implemented base models, ensuring compatibility with custom datasets. Dataset Preparation and Fine-Tuning: Prepared and captioned datasets for model training, tailoring datasets to various styles and categories. Conducted fine-tuning using Dreambooth and LoRa techniques, iterating with multiple parameters (epochs, steps, learning rates) to optimize model performance. Implemented advanced fine-tuning techniques, including selecting optimal samplers and steps to enhance image generation quality. Model Training and Evaluation: Trained and fine-tuned Stable Diffusion models on custom client datasets, improving AI-generated image reliability and customization. Evaluated model performance using multiple functional sprints and refined generation processes based on results. Client Collaboration and Workshops: Collaborated closely with a team of four engineers to meet client requirements and deadlines, enhancing team efficiency and client satisfaction. Conducted client workshops to gather functional and technical requirements, utilized prompt engineering, and provided training for non-technical users on platform usage. Technology Management: Managed the RUNPOD platform, ensuring hardware and software resources met project demands, leading to improved project performance and scalability.

Pedagogical Assistant for NVIDIA courses and certified workshops: "Fundamentals of deep learning" "Generative AI with Diffusion Models"

Summer Internship in Email Sentiment Analysis Developed data processing pipelines for a variety of email formats using automation techniques. Initiated the creation of a highly precise sentiment analysis model utilizing Deep Learning techniques, achieving 95% email classification accuracy.

MASSAI Link : https://massai.esprit.tn Pedagogical Assistant for two NVIDIA courses and two certified workshops: "Sequential data modeling with recurrent neural networks in Keras" "Introduction to physics-informed machine learning with Modulus" "Acceleration of end-to-end data science workflows" "Creation of real-time AI video applications"

Summer Internship in Recommendation System Designed Deep Learning models with TensorFlow and OpenCV to classify glasses and facial structures, achieving a 95% accuracy rate. Integrated the models into a real-time recognition tool on a sales platform.

MASSAI Link : https://massai.esprit.tn Collaborated closely with esteemed instructors to enhance participants' understanding of complex topics in machine learning and AI. Offered valuable insights and assistance to participants, helping them overcome challenges and gain proficiency in specialized areas of AI and deep learning. Contributed to administrative tasks to optimize workshop execution, exemplifying attention to detail and organizational aptitude. Assisted in organizing and coordinating workshop logistics, ensuring a seamless and engaging learning environment. Pedagogical Assistant in 4 certified NVIDIA workshops: "Fundamentals of deep learning" "Fundamentals of accelerated computing with CUDA C/C++" "AI applications for anomaly detection" "AI applications for predictive maintenance"