Aimen Boukhari

Exploring unseen patterns in vision and text, self-supervised learning, neuro-symbolic AI, and machine reasoning Models, Agentic AI

About

Fourth-year Computer Science student at ESI Algiers with a focus on deep learning, computer vision, and generative AI. Actively pursuing research in self-supervised learning, world models, and agentic AI systems. Experienced in implementing cutting-edge architectures and reproducing research papers to advance both theoretical understanding and practical applications.

"For me mathematics is just an instrument I enjoy playing. I don't care if the problem is important or not, as long as it's an interesting puzzle. Some of the best things I've done have been completely unimportant."

— Freeman Dyson

Publications

Predict and Reconstruct: Joint Objectives for Self-Supervised Language Representation Learning
Preprint, ResearchGate, 2025
Under submission

Featured Projects

I-JEPA Embedding Inversion Attack
2025
PyTorch Self-Supervised Learning Security Representation Learning
Research project investigating inversion attacks on I-JEPA embeddings to analyze privacy and security vulnerabilities in self-supervised world models. Implemented embedding inversion techniques to reconstruct input signals from learned representations, demonstrating that predictive self-supervised architectures can be vulnerable to representation inversion attacks.
PaperLens
2025
LangGraph Python LLMs Vector Search
An intelligent research discovery platform that automatically crawls ArXiv papers, generates multimodal embeddings, and delivers personalized recommendations through advanced vector search and machine learning-driven user preference modeling.
TravelAgent (TravelMate AI)
2025
LangGraph Python LLMs Agentic AI
An agentic AI system for travel planning using LangGraph and LLM orchestration. Integrates external APIs to provide personalized travel recommendations and itineraries.
I-JEPA Architecture (Inspired by Yann LeCun)
2025
PyTorch Self-Supervised Learning Vision
Implementation of an early version of the I-JEPA architecture focusing on predictive representation learning. Experimented with self-supervised approaches for vision and reasoning tasks.
Plant Disease Detection using SSD300
2024
PyTorch Object Detection Computer Vision
Implemented SSD300 object detection model from scratch and trained it on real-world plant disease images in the Zindi Competition.
Variational Autoencoders (VAEs) from Scratch
2024
PyTorch Generative Modeling Representation Learning
Implemented VAE architecture from scratch and trained on MNIST. Explored representation learning and generative modeling for large-scale vision tasks.

Blog & Writing