AI Engineer · Software Engineer

Building AI-powered products and production software with Python and TypeScript.

Fourth-year student at ENSIA — National School of Artificial Intelligence, Algiers

I build complete systems, not demos — from data pipeline to trained model to a working product someone can actually use. My work spans healthcare, CRM, and e-commerce, mostly for the Algerian market.

On the ML side, I've built a wildfire risk model from a dataset I collected myself, and trained YOLOv8 for both object detection and adversarial attacks. On the software side, I built and shipped Tajdeed, a full platform for certifying second-hand electronics, across desktop, backend, and Android.

Right now I'm deepening my AI security work and starting a serious LLM/RAG project to round out my AI engineering skills.

AI / ML:

Backend:

Frontend / Desktop:

Data & Infra:

3-month internship at Mobilis. I built the Flutter mobile app and the Django backend (JWT auth, PostgreSQL, Docker) for an internal CRM system. The mobile app repo belongs to the company, so only the backend is linked here. Backend Repo · Demo Video Flutter Django PostgreSQL Docker

I designed and built the desktop app for a platform that lets Algerian phone shops diagnose, certify, and resell second-hand devices. The platform is split into three parts — a desktop app, a Node/Prisma backend, and an Android diagnostics companion — and I engineered the shipped features: device diagnostics, digital certificates, a marketplace, and a payment/settlement flow. Website · Desktop App Releases Electron Next.js Node.js Prisma PostgreSQL

I collected a dataset of 121,869 labelled cell-days across 14 fire seasons directly from NASA FIRMS and Open-Meteo APIs, then trained a LightGBM model to predict wildfire risk. Mid-way through, I caught a label-leakage bug that was inflating my results by 21% — fixing it and re-validating with year-grouped cross-validation got me to a realistic 0.559 PR-AUC, plus base-rate-corrected risk bands for deployment. Repo · Dataset on Kaggle Python LightGBM Geospatial Data Engineering

I built a desktop app for medical practices that manages patients and consultations, and engineered an AI assistant on top of it that pulls a patient's full history into context so it can support clinical decisions with Gemini. Demo Video Electron Next.js SQLite Gemini API

I built a web app that parses resumes (PDF parsing pipeline) and generates structured, real-time feedback using Claude Sonnet 5. Repo · Live App React Claude API PDF Parsing

I modeled drone delivery over Algiers as a constrained vehicle routing problem (battery, payload, no-fly zones) and implemented and benchmarked three different solvers — branch & bound, a genetic algorithm, and simulated annealing — against each other. Repo Python Optimization Genetic Algorithm

• Algeria Weather — Desktop weather app covering 157 Algerian cities in English, French, and Arabic, no API key required.
• Pixel — Adversarial Attacks — White-box (FGSM) and black-box (patch) attacks against a YOLOv8 aircraft detector. Project site
• Appliance Power Classification — Classified 10 home appliance types from power-usage traces, 97% accuracy on a course competition.
• Brent Oil Forecasting — Tested whether ARIMA models beat a naive random walk on monthly Brent crude prices — none did.

• Agentic AI
• Advanced Deep Learning
• Cloud Computing
• Production AI Engineering
• System Design

AI Engineering · AI Product Development · Software Engineering