A selection of professional, personal, and open-source projects spanning applied AI, industrial systems, and software development.
Developed an agentic AI application that uses large language models, LangGraph, and the Model Context Protocol to provide natural-language access to complex industrial data systems. The agent coordinates tools and data retrieval to help users discover assets, inspect metadata, retrieve sensor measurements, analyze historical trends, and investigate operational conditions.
Developed reusable skills for OpenAI Codex that enable operations teams to access and analyze complex data systems using natural language. Deployed in an industrial environment, the skills encapsulate domain-specific workflows and execute Python tools against internal APIs, allowing users to retrieve sensor measurements, explore asset metadata, analyze historical trends, and investigate operational performance.
Designed and deployed a perception system for autonomous industrial trucks that processes data from 8 cameras, 9 radars, and 2 LiDAR sensors. The system combines deep learning and sensor fusion to support object detection, scene understanding, and safe navigation in complex operating conditions.
Developed and deployed an image segmentation solution to estimate ore particle size distribution from industrial images. The system uses the Segment Anything Model on edge devices to support quality control by monitoring particle sizes and identifying oversized ore that could disrupt or damage downstream processing equipment.
Applied reinforcement learning within a digital twin to learn control strategies for an industrial process. The model was trained to adjust operating conditions and regulate system temperature around defined target setpoints.
Developed machine learning models using industrial time-series data to estimate equipment remaining useful life and identify emerging degradation patterns, supporting proactive maintenance planning and reducing the risk of unplanned downtime.
Designed a real-time recommendation system that compares current operating conditions with similar historical process states. The system identifies past control actions associated with better outcomes to help reduce process variability and improve operational stability.
Built an open-source MCP server that connects AI clients to Morocco's data.gov.ma portal. The server exposes typed tools for searching and retrieving datasets, resources, organizations, and groups through the CKAN API, with validation, error handling, automated tests, and Docker support.
Built and launched verity.ma, a website for sharing and discovering customer experiences with companies in Morocco. Developed the platform end to end using Django for the backend and React for the frontend, including the APIs, database, deployment infrastructure, and content management workflows.
