Work
Project 01 / 06
@ Metavizor Labs
Agentic platform turning raw data into fine-tuned models — automatic model selection, LoRA/QLoRA training and evaluation for text and image tasks.
Project 02 / 06
Mobile app to log meals via food-photo recognition, chat with an AI health coach, track nutrition & health metrics and get personalized recommendations.
Project 03 / 06
Full-stack AI platform for market analysis, forecasting, anomaly detection, portfolio management and multi-agent orchestration on the Tunisian stock exchange (BVMT).
Project 04 / 06
Machine-learning system that analyzes network traffic to detect anomalies and intrusions.
Project 05 / 06
E-commerce platform with personalized recommendations, a real-time support chatbot and an admin dashboard for products, orders and users.
Project 06 / 06
Credit-risk model to identify loan defaults and analyze the key customer risk drivers.
Flagship builds
Summaries of my two most ambitious projects. Full case studies are on the way.
An agentic platform, built as my end-of-studies project at Metavizor Labs, where a user drops in raw data and receives a fine-tuned, evaluated model with minimal input — across both text and image tasks.
Keywords
Agentic workflows, LangGraph, parameter-efficient fine-tuning, LoRA, QLoRA, model selection, automated evaluation.
A full-stack AI platform for the Tunisian stock exchange (BVMT) combining forecasting, anomaly detection and portfolio tools, coordinated by a team of specialised agents.
Keywords
Time-series forecasting, anomaly detection, portfolio management, multi-agent orchestration, BVMT.
Build lab
Planned builds — not finished yet, so no links. Here's what I'm exploring next.
Differential privacy for clinical models
Training diagnostic models on sensitive health data while bounding what the model can leak about any individual patient.
Hybrid quantum–classical models
Benchmarking variational quantum classifiers against classical baselines on small, structured datasets.