Nourhene Methlouthi

Autonomous Medical Report Generation (Azure AI, LangChain, Python, OpenAI) Designed and orchestrated autonomous medical AI agents on Azure AI using LangChain to manage dialogue, reasoning, and workflow execution. - Built an end-to-end pipeline on Azure OpenAI from ASR (Automatic Speech Recognition) of doctor dictations → error correction models → retrieval-augmented generation (RAG) for contextual enrichment. - Indexed a large medical terminology/database in a vector DB (FAISS / Pinecone) to enable semantic similarity search and accurate alignment of transcribed text with standardized terms. - Implemented LLM-powered template filling, where extracted entities (pathologies, exams, patient data) were automatically mapped into structured medical report templates. Results: Delivered a system that could generate standardized, accurate, and ready-to-use medical reports from noisy doctor dictations, significantly reducing manual transcription and errors.

Delib: AI-Powered Government Deliberations Processing (RAG, OpenAI, Web scraping, Next.js, Python, PostgreSQL, Elasticsearch) - Designed an end-to-end data pipeline to crawl government websites for deliberation PDFs, extracting structured information using OCR (Pytesseract). - Applied OpenAI prompting to identify and extract investment project details (date, type, city, budget), effectively transforming unstructured documents into actionable data Architected PostgreSQL database and Elasticsearch index to store, query, and retrieve project information; automated nightly updates using Celery + cron jobs with multithreading/multiprocessing for high performance. - Developed a Next.js dashboard to visualize and interact with stored data, including secure login (JWT/Google/LinkedIn), advanced search (autocomplete, filters, favorites), push notifications, and PDF/Excel exports. Results: Achieved 83% real-time detection and structured storage of new investment projects, enabling faster decision-making and analysis.

- Developed and implemented object detection and tracking algorithms for autonomous robots to assist with the installation of fiber optic cables in underwater tanks. - Preprocessed point cloud data from ToF cameras and a 360-degree 3D LIDAR using Point Cloud Library (PCL) and Principal Component Analysis (PCA) libraries, employing cropping techniques to filter clouds and form datasets for object tracking. - Developed advanced object detection and classification algorithms using Random Forest models, meticulously optimizing the algorithm with GridSearch for hyperparameter tuning, ensuring superior performance. -Optimizing tracking algorithms YOLOv7-tiny by reducing model size and computational complexity through techniques such as model pruning, quantization and utilizing efficient layer structures. - Engineered a sophisticated position and orientation algorithm to accurately predict the object orientation using Iterative Closest Point (ICP) alignment with orthogonal Procrustes initialization, enhancing object tracking precision. - Utilized linear algebra methods to calculate the final position and matrix of orientation for each object detected, ensuring precise spatial understanding and navigation. - Integrated algorithms seamlessly into the Robot Operating System (ROS) for real-time execution on GPU targets running Ubuntu Linux, ensuring efficient and reliable operation in dynamic environments. - Spearheaded the optimization of perception for TwinswHeel logistics drones within a 10-meter radius, enabling drones to navigate dynamically and reach their targets in real-time, significantly enhancing logistics efficiency.

- Collect historical data on maintenance record and failure logs and use Apache Spark for Data cleaning, transformation and integration from various sources. - Extract relevant features from raw sensor data such as vibration levels, temperature, pressure and operational hours. - Use Spark MLlib to develop and train a Random Forest model on the preprocessed data. - Use Spark Streaming for real time dara processing and model prediction - Implement a CI/CD pipeline for model deployment and updates

- Designed and implemented a license plate detection system using the YOLOv7 model on a Jetson Nano to capture and detect license plates in real time. - Integrated Optical Character Recognition (OCR) to accurately extract license plate numbers from detected plates. - Preprocessed images to enhance OCR accuracy through resizing, thresholding, and noise reduction. - Created a real-time monitoring system to log and track vehicle activities using an Arduino microcontroller for data collection and transmission. - Set up a MySQL database to securely store client data, vehicle information, and activity logs..