Nadhir Benhalima

I've delivered industrial AI systems for STEG, Tunisia's national electricity utility, from SCADA data acquisition through production decision-support interfaces. My predictive-maintenance programme for turbo-alternator partial discharges achieved R² ≈ 0.97 with Physics-Informed Neural Networks, compared with a 0.49 analytical baseline.

At STEG, I built pipelines across five years of SCADA records, benchmarked machine learning and deep learning models, and deployed MLOps workflows for versioning, experiment tracking, monitoring, explainability, and continuous retraining. I also shipped a React and FastAPI real-time dashboard with WebSocket latency under 100 ms and managed planning, stakeholder reporting, and delivery milestones.

As a freelance AI Research Engineer at ReDCAD Laboratory, I'm designing a privacy-preserving federated learning framework across Reddit, LinkedIn, and Amazon. I'm fine-tuning LLMs locally with LoRA, aggregating updates with FedAvg, and building a targeted marketing recommendation engine delivered through REST APIs and an evaluation dashboard.

Alongside AI research, I build full-stack applications, ETL pipelines, NLP components, predictive models, and containerised services using React, Next.js, FastAPI, Node.js, Docker, and AWS. I'm fluent in Arabic, English, and French, and available for remote roles internationally.