Skip to content

Projects

From a scanner running in French retail to a deep learning timetable planner.

Featured

Limpscanner

Production browser based 1D and 2D barcode scanner. An Angular library using WebGL for GPU image processing, ZBar via WebAssembly, and a multi stage pipeline with region of interest localization. Running at Castorama.

  • Angular
  • TypeScript
  • WebGL
  • WASM
  • Emscripten
  • Web Workers

Automatic timetable generation for a university, framed as sequential decision making under hard constraints. A PyTorch policy places one assignment at a time, masked to only the moves the constraints allow, so every schedule it builds is valid by construction. Reinforcement learning then tunes that policy against the quality of the whole timetable rather than any single slot.

  • Python
  • PyTorch
  • Reinforcement Learning
  • Deep Learning
  • Optimisation

Heart Attack Prediction

Supervised classification on medical data, comparing logistic regression, random forest, and XGBoost with hyperparameter tuning and cross validation. Scored on F1 and ROC AUC, with attention to fairness, feature normalisation, and the privacy that health data demands.

  • Python
  • scikit-learn
  • XGBoost
  • Classification

A medical chatbot built on a forward chaining inference engine. A knowledge base of IF THEN rules reasons over reported symptoms to narrow toward a probable diagnosis, built on the AIMA framework.

  • Python
  • Expert System
  • Inference Engine
  • AIMA

Social and community app for ESTIN students: a feed for campus news, events, and discussion, with a shared digital library for course files.

  • Android
  • Kotlin
  • Firebase

Mobile app for Algerian baccalaureate exam preparation, giving students structured access to past papers and revision material.

  • Android
  • Kotlin