Mayssa Jaziri

• Analyzed traffic data from an intersection in Istanbul to identify trends, peak hours, and significant patterns, integrating the results into the Kentyou mobility solution. • Conducted a comparative analysis of forecasting models and implemented an LSTM model to predict vehicle count every 15 minutes with a 5-minute window. • Developed a predictive microservice to forecast traffic volume and queue length, optimizing traffic light cycles. • Built an MLflow pipeline integrating: - Data preprocessing - Model training and evaluation - Visualization of results • Used MLflow to track experiments with different hyperparameters and model comparisons. • Reviewed literature on LLMs and MLLMs in order to select and adapt these models to generate descriptive text for traffic charts. • Implemented and performed unit tests on a service dedicated to analyzing traffic charts at specific junctions, generating automated monthly PDF reports that highlight and explain key chart components using MLLMs; and containerized the service using Docker for deployment. • Implemented a RAG system to answer customer questions using Kentyou’s internal knowledge base, relying on a chatbot interface developed in Streamlit. • Upgraded the initial RAG system into a distributed solution based on MCP servers to improve the performance and scalability of the solution.

• Contributed to the implementation of a fully automated machine learning training workflow within a P2P architecture, maximizing efficiency and scalability while reducing operational complexities. • Designed and implemented robust security measures for the architecture and adapted the solution to Serverless computing, addressing challenges related to timeout and lack of state. • Achieved a high accuracy rate exceeding 83% against three different simulated attacks by varying the number of malicious entities inside the training network. • Contributed to the writing of a scientific paper accepted at the IEEE International Conference on Software Quality, Reliability, and Security (QRS).

• Developed a CV analysis microservice to extract key details about the candidate, including contact information, educational background, work experience, and skills associated with each work experience. • Acquired a microservice capable of capturing up to 1908 distinct skills. • Containerised the microservice using Docker to facilitate its integration with the candidate matching microservice for a specific job description.

• Collected and preprocessed images from various locations along the Tunisian coast. • Extracted information about the Tunisian coastline for three distinct years from the acquired images. • Designed and developed an advanced erosion monitoring system on the Tunisian coast using remote sensing techniques, geographic information systems (GIS), and predictive modelling.

• Designed and developed the INAYA application, a website dedicated to Tunisian ophthalmologists. • Implemented an immersive chat feature using sockets, promoting dynamic interaction. • Maintained close communication with the client throughout the internship to proactively address specific needs.