Chaouki HJAIJI

– Developed an LLM-powered log analysis system for 5G networks using open source LLMs like Llama3.2. – Worked with curated datasets for AI-driven log analytics research like BGL, HDFS,.. with more than 11M logs. – Processed rows of logs and stored them in Appache Arrow format for 80% of efficiency in its treatment. – Collaborated with MonacoTelecom to implement the approach in its environments with more than 50M logs. – keywords: LLMs, Finetuning, PEFT, Ollama, vLLM, Log Analysis.

– Finetuned open source LLMs to enhance the technical support experience of the company's customers. – Conducted supervised fine-tuning of open source LLMs like Llama-7B on customer datasets. – Used Parameter Efficient Finetuning, LoRA and quantization techniques for faster and efficient finetuning. – Evaluated system performance using human feedbacks. – keywords: LLMs, Unsloth, Pytorch, PEFT, NLP.

– Built a classifier for brain tumor image classification with 90% precision to evaluate the impact of data augmentation. – Generated synthetic medical images using GAN-based models such as Deep Convolutional GAN. – Injected varying percentages (15%, 35%, 50%, and 80%) into the original dataset. – Evaluated the approaches using the Brain Tumor Detection MRI dataset, containing 3,000 MRI images of brain tumors. – keywords: GANs, Pytorch, Image Processing.

– Worked on a sustainable energy management project, including the development of smart metering tools such as the electrical meter Voltix. – Analyzed the state of the art in electrical energy disaggregation and identified relevant approaches. – Implemented various models to predict the energy consumption of individual appliances with an accuracy of 92%. – Utilized a diverse range of publicly available datasets, such as the REDD dataset, which contains 13 million samples of electrical energy measurements. – keywords: Data Preprocessing, Feature Engineering, Scikit-Learn, Numpy, Pandas, Seaborn.