🧐 About Me

Hi there! I’m Ayoub El Khaiari, a recent graduate with a Master’s in Advanced Machine Learning and Multimedia Intelligence from Faculté des Sciences de Fès. I am currently conducting research in diverse areas of computer science, focusing on transforming ideas into intelligent systems and bridging research with real-world applications.

🚀 AI & Data Science Engineer

💡 Transforming Ideas into Intelligent Systems

🧠

💬

👁️

✨

🔬 From research to real-world applications

Open to: AI Engineering & Research, Software Engineering & Security, GenAI and Industrial AI and Software.

Research Interests: My research interests lie in low-level AI, where mathematics plays a central role in understanding and optimizing model behavior. I’m also deeply interested in AI for Healthcare, software security (particularly malware detection), and Generative AI, including Small Language Models (SLMs), knowledge distillation, and Inverse Reinforcement Learning.

🚀 Currently, I’m working on a project that integrates Large Language Models (LLMs) and Vision-Language Models (VLMs) for time series forecasting in an interpretable, language-based framework.

Please feel free to contact at ayoub dot elkhaiari at yahoo dot com for collaboration.

🔥 News

2025.10: OpenAI releases GPT-5, introducing multi-agent reasoning and memory-persistence features (Major Update)!
2025.10: DeepMind publishes “Self-Taught Optimizers”, showing neural networks can learn gradient descent-like rules from scratch.
2025.09: Meta unveils LLaMA-4, with 2 trillion parameters and open multimodal APIs (Open Release)!
2025.09: Apple announces Apple Intelligence on-device SLMs for iPhone 16, integrating privacy-preserving GenAI.
2025.09: Anthropic’s Claude 4.1 surpasses GPT-4 Turbo in reasoning and safety evaluation benchmarks.
2025.08: Google DeepMind introduces Gemini 2, a unified multimodal reasoning model integrated into Workspace and YouTube.
2025.08: The EU officially enforces the AI Act, requiring interpretability and energy-efficiency disclosures for large models.
2025.07: NVIDIA debuts Blackwell Ultra GPUs, offering 3× faster mixed-precision training performance (Hardware Leap)!
2025.06: MIT and Stanford researchers release Self-Evolving Agents, a benchmark for long-horizon autonomous AI.
2025.05: Microsoft integrates Copilot across Windows, Excel, and VS Code with cross-context memory and speech understanding.

📝 Publications

Oct 2025

Learning Heuristic Solutions for the Traveling Salesman Problem Using Graph Attention Networks and Reinforcement Learning
Author, Research Report - Learning Heuristic Solutions for The TSP, Fez, Morocco.

Learning Heuristic Solutions for the Traveling Salesman Problem Using Graph Attention Networks and Reinforcement Learning. Proposed a deep learning framework that leverages graph attention networks combined with REINFOREC algorithm in reinforcement learning to learn efficient heuristic solutions for the Traveling Salesman Problem. Focused on optimizing route selection by capturing node dependencies and sequential decision-making, improving solution quality and generalization over unseen graphs, while reducing computation compared to classical heuristics.View Medium

Sept 2025

Multi-modal Depression Estimation using Adaptive Attentional Fusion Networks
Co-Author, Research Report - Multi-modal Depression Estimation, Nancy, France

Multi-modal Depression Estimation using Adaptive Attentional Fusion Networks. Proposed a deep learning framework combining multi-modal data through adaptive attention-based fusion. Focused on robust estimation of depression severity from heterogeneous inputs, improving interpretability and accuracy. Achieved more than 40 downloads: View Report

Aug 2025

MISB-HMM RUL for Multiple Deterioration Modes Systems
Co-Author, Research Report, Nancy, France

Multi-inter-Switch Branches HMM for Multi-mode Degradation, statistical and probabilistic modeling for systems under multiple deterioration, estimating the Remaining Useful Life. Achieved more than 130 downloads: View Report

🎖 Honors and Awards

2025.10 Selected and successfully passed the interview for the Master’s graduation research internship, among top participants.
2023.08 Selected for the Master’s program MLAIM (Advanced Machine Learning & Multimedia Intelligence) — first promotion of this innovative and excellence-driven program.
2023.05 Received the Best Graduation Award for the bachelor’s project “Smart Home”, integrating Robotics, Artificial Intelligence, and Android Development, as part of the Bachelor’s degree in Mathematics and Computer Science.

💻 Experiences

2025.10 - 2025.10, Faculty of Sciences, Technology & Medicine, University of Luxembourg, Belval, Luxembourg, Research Contributor, Luxembourg
_{Contributed to the development of a High-Performance Computing (HPC) e-learning platform aimed at advancing computational education. Focused on optimizing user interaction and performance modules using Advanced GenAI models, that made learning complex tasks in HPC easier.}
2025.03 - 2025.08, Institut Jean Lamour, University of Lorraine, Nancy, France, AI Research Intern, France
_{Developed AI models for synchrotron XRD analysis, generating synthetic diffraction data with less than 3% phase length deviation. Applied deep learning architectures (CNNs, Transformers, LSTM-Attention) to predict steel phase fractions, and used XAI techniques (SHAP, LIME) to interpret alloy composition–phase relationships.}
2024.12 - 2025.04, Outlier Remote – Project Cypher_RLHF, AI Trainer, Remote
_{Created and refined prompts to probe LLM weaknesses in reasoning and instruction-following. Evaluated model outputs for accuracy, writing quality, and cultural relevance, and ranked responses to guide RLHF fine-tuning. Enhanced robustness of instruction-tuned models through prompt engineering and adversarial testing.}
2024.07 - 2024.08, RMA ARZ Assurance, Rabat, Morocco, AI Engineer Intern, Morocco
_{Extracted insurance documents using OCR and PySpark pipelines, and integrated GraphRAG with late chunking (via Jina.AI) to improve reasoning and reduce hallucinations and even added some custom chunking. Built a HybridRAG system for structured/unstructured querying simple RAG & GraphRAG, a text-similarity recommender, and multilingual agents for language detection, contextual translation, and recommendation using PEFT fine-tuning and prompt engineering.}
2022.02 - 2024.06, Freelance Web Developer, Remote
_{Designed and deployed full-stack web applications using MERN, MEAN, Laravel, Django, and Spring Boot. Specialized in API integration, software architecture, and database design (MongoDB, MySQL, GraphQL). Delivered robust, client-oriented digital solutions with responsive interfaces and scalable backends.}

🧠 Projects

🤖 Hybrid & AI Systems

🚀 AI Resume Matcher — Developed an AI resume screener with CI/CD pipeline; FastAPI backend and Streamlit frontend in separate Docker containers.
🧰 Tools: FastAPI, Streamlit, Docker (Ubuntu), CI/CD, Sentence-Transformer.
🤖 Human-in-the-Loop RL for Sentiment Prediction — Developed a TRPO-based agent using PyTorch and Gym to classify sentiment from BERT embeddings. Integrated adaptive re-training from user feedback with safe policy updates to reduce bias and enhance robustness.
🧰 Tools: PyTorch, Gym, TRPO, BERT, Human-in-the-Loop Learning.
🏠 Smart Home with Speech Recognition — Designed a smart home prototype with voice-controlled devices via Arduino–Java serial communication.
🧰 Tools: Arduino, Java, Speech Recognition, Android Studio.
🧭 PPO-Agent for Graph-Based Community Detection — Modeled community detection as a sequential decision task using PPO in a custom Gym environment.
🧰 Tools: PyTorch, Gym, PPO, Node2Vec, NetworkX.
🤖 AI Agents — Built autonomous multi-tool LLM agents (DeepSeek-R1) with Streamlit interface for interactive reasoning.
🧰 Tools: Streamlit, Python, LLM Integration.
🧩 Reinforcement Learning with Graph Attention Networks for Solving the Traveling Salesman Problem (TSP) — Combined Graph Attention Networks with Reinforcement Learning to approximate optimal TSP routes, demonstrating promising results on random graphs and potential for real-world scalability. 🧰 Tools: PyTorch, Graph Attention Networks (GAT), Reinforcement Learning, NetworkX, Matplotlib.

🛰️ Computer Vision

🖼️ Image Processing Application — Applied Gaussian, Median, Sobel, and Canny filters for enhancement and edge detection. Used FFT and Hough Transform for frequency and shape analysis.
🧰 Tools: OpenCV, NumPy, FFT, Image Filtering.
🚦 TrafficSignNet — Built a traffic sign recognition model using ResNet50 with Grad-CAM and LIME for explainable AI. Achieved high accuracy on GTSRB with detailed performance visualizations.
🧰 Tools: TensorFlow, ResNet50, Grad-CAM, LIME.
🧨 Adversarial GANs — Built an attack-defense framework using GANs and chaotic encryption to enhance cybersecurity robustness. 🧰 Tools: PyTorch, Chaotic Encryption, GAN.
🛰️ SatEye — Built an IRL-based object detection model for satellite imagery using MaxEnt and ResNet-50.
🧰 Tools: PyTorch, ResNet-50, MaxEnt IRL, Convex Optimization.
🧩 Image Segmentation Using Multi-Head Attention & K-means — Built an image segmentation model combining a multi-head attention encoder for feature extraction and K-means clustering for grouping. Visualized color distributions in 3D for clear segmentation insights.
🧰 Tools: PyTorch, Attention Encoder, K-means, OpenCV, Matplotlib 3D.
👟 Virtual Try-on System for Footwear (3D & AR) — Captured RGB-D images via Kinect for markerless 3D foot tracking and virtual shoe fitting. Used ICP for model registration and optimized transitions for 24 FPS real-time performance.
🧰 Tools: Microsoft Kinect, C++, Open3D, ICP, AR Tracking.
🌫️ Denoising Diffusion Model (DDPM) for Image Generation — Implemented DDPM with optimized noise scheduling and sinusoidal embeddings for U-Net-based denoising. Achieved stable generation on CIFAR10 and Sprite datasets.
🧰 Tools: PyTorch, DDPM, U-Net, Sinusoidal Embeddings.
🚦 Real-Time Object Detection & Tracking — Applied Mask R-CNN and YOLOv8 for real-time object detection and DeepSORT for multi-object tracking with trajectory visualization and speed estimation.
🧰 Tools: PyTorch, YOLOv8, Mask R-CNN, DeepSORT, OpenCV.
🏗️ 3D Scene Analysis & Point Cloud Classification — Built a 3D reconstruction pipeline using SIFT and PointNet.
🧰 Tools: PyTorch, Open3D, SIFT, PointNet.
🎮 3D Transformations & Projections — Implemented 3D transformations (translation, rotation, scaling) and projections (Cavalier, Cabinet, Perspective).
🧰 Tools: C++, Linear Algebra, Computer Graphics.
🤝 Federated Learning with Flower — Created a federated neural network on MNIST enabling decentralized collaborative training with strong privacy guarantees.
🧰 Tools: PyTorch, Flower, Federated Learning.

🧠 NLP & LLMs

🗞️ FLAN-T5 Summarization — Fine-tuned FLAN-T5 using PEFT (LoRA) and RLHF (PPO) for dialogue summarization with toxicity control.
🧰 Tools: PyTorch, Transformers, Perspective API, ROUGE.
🧮 NLQ-SQL — Built a natural language to SQL converter using Gemma LLM and schema extraction for Oracle databases.
🧰 Tools: OLLAMA, Gemma LLM, OracleDB.
🐦 Real-Time Twitter Sentiment Analysis — Built a live tweet sentiment pipeline with BERT, PySpark, and MongoDB, fully automated with ETL and real-time PowerBI dashboards.
🧰 Tools: PySpark, BERT, TextBlob, Docker, MongoDB, PowerBI.
📝 Spelling Correction System — Built a probabilistic spelling corrector using edit distance with dynamic programming and frequency-based ranking from Shakespeare’s corpus.
🧰 Tools: Python, NLP, Dynamic Programming.
🌐 Machine Translation — Developed EN–FR translation systems (LSH-based and Seq2Seq with attention).
🧰 Tools: PyTorch, LSTM, BLEU Score, LSH.

–>

📖 Educations

^🎓 2023.09 - 2025.07, Master in Advanced Machine Learning & Multimedia Intelligence, Sidi Mohammed Ben Abdellah University, Faculty of Science, Fez, Morocco.
_{Key courses: Image Processing · Data Governance · Search Algorithms · Metaheuristics · Markov and Stochastic Processes · Computer Graphics · Big Data · Oracle Databases · Business Intelligence (BI) · Operations Research · Machine Learning and Deep Learning · Data Mining · Data Engineering · Video Analysis · Speech Recognition · Smart Cars · Augmented Reality · Image Mining · Text Mining · 3D Scenes analysis · Deep Reinforcement Learning · Multimodal GenAI · Medical Imagery}
^🎓 2019.09 - 2023.06, Bachelor in Mathematics & Computer Science, Sidi Mohammed Ben Abdellah University, Faculty of Science, Fez, Morocco.
_{Key courses: Algebra · Analysis · Numerical Analysis · Probability · Algorithms · C · C++ · Java · Web Development (HTML, CSS, JS, PHP) · Operating System Programming · Networking · Network Programming · Databases · Object-Oriented Programming (OOP) · UML · Project Management}
^🎓 2019.09 - 2022.06, Associate’s Degree in Mathematics & Computer Science, Sidi Mohammed Ben Abdellah University, Faculty of Science, Fez, Morocco.
_{Key courses: Algebra · Analysis · Numerical Analysis · Probability · Algorithms · C · Web Development (HTML, CSS, JS, PHP) · Operating System Programming · Databases · Object-Oriented Programming (OOP) · UML · Project Management · Thermodynamics · Optics · Algorithms}

🎙 Miscellaneous

Travel

I enjoy traveling and exploring new countries and cultures. Some of my favorite destinations include Switzerland, Luxembourg, Belgium, Turkey, France, and Germany. Each trip inspires me to see the world from different perspectives.

France, Nancy

Switzerland, Bale

Turkey, Istanbul

Luxembourg, Belval

France, Saint-Michel

Turkey, Antalya

Sports

I’m passionate about martial arts, especially Taekwondo and Kickboxing, which I have practiced for years. These sports have taught me discipline, focus, and perseverance.

Teakwondo