Redouane Hadni

Development of an AI diagnostic agent for industrial machines, enabling field technicians to resolve incidents autonomously without human intervention. Agentic architecture in Python with an intelligent LLM router, dynamic source selection, automatic citation of reference documents. F1-score of 0.88 on test dataset. Deployment on server via Docker; Chainlit interface; automated Google Drive synchronization.

Development of a full-stack end-to-end AI application for automatic extraction of key information from administrative documents (sick leave certificates, expense reports, etc.). Hybrid architecture combining OCR and LLM (GCP Vertex AI) deployed on a Python/FastAPI backend with an Angular frontend. Achieved 97% accuracy on sick leave documents, with an average response time of 3 seconds and a cost of €0.20 per 1,000 documents.

- Built a high-frequency database covering 7 consecutive days for multiple cryptocurrencies (1.8 B ticks): cleaning, key-stat computation, microstructure-noise quantification, and exploratory visualisations using Python, Pandas, and Matplotlib for publication and internal use.- Designed and developed a Python reinforcement-learning library for finance: supports multi-asset training and back-testing on real data, flexible feature selection for the state space, and optimized training pipelines.