Skip to content
View elymeraj's full-sized avatar

Block or report elymeraj

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
elymeraj/README.md

Eldis Ymeraj

AI & Machine Learning Engineer · Computer Vision · NLP · Data

Portfolio · LinkedIn · Email

English · Français


English

About me

AI & Machine Learning Engineer with hands-on experience in developing, evaluating, and analyzing machine learning and deep learning models. Currently working at Airbus on Explainable AI for computer vision, with additional experience in NLP, data science, and applied machine learning.

Professional Experience

Artificial Intelligence Intern - Trustworthy AI · Airbus SAS
Toulouse, France · March 2026 – September 2026

  • Adapted an Explainable AI method to a multi-scale YOLO-NAS Pose model using PyTorch to identify the visual features influencing model detections.
  • Developed an experimental protocol to extract, select, and analyze visual representations, assessing their quality, stability, and ability to preserve information relevant to detections.
  • Quantitatively evaluated detection fidelity using detection and spatial matching metrics, achieving 90.5% precision and a PR-AUC of 0.776.
  • Analyzed model behavior across different runway distances to identify the most influential visual information and how it evolves throughout the approach.

Data Scientist Intern · Responsibio
Caen, France · May 2025 – August 2025

  • Collected and prepared mouse vocalization recordings and associated metadata to build a dataset for animal welfare classification.
  • Developed an audio classification pipeline using mel-spectrograms, feature extraction, PCA dimensionality reduction, and a linear SVM classifier.
  • Evaluated the model using 5-fold cross-validation, reducing 2048 features to 58 while retaining 86% of the variance and achieving approximately 79% correct predictions.

Featured Projects

PythonPyTorchYOLOv8OpenCVByteTrackHomography

  • Created and annotated datasets for players, goalkeepers, referees, the ball, and court keypoints, then trained specialized YOLOv8 detection models.
  • Developed a video tracking pipeline combining ByteTrack, ball filtering and interpolation, with homography-based projection of player and ball positions onto a 2D court.

PythonPyTorchHugging Face TransformersCamemBERTPandas

  • Developed an NLP system to automatically identify episodic and semantic memory spans at token level in French clinical narratives.
  • Fine-tuned and compared CamemBERT, ModernCamemBERT, and DeBERTa using patient-level cross-validation to evaluate generalization while preventing data leakage.

Other Projects

Technical Skills

  • Programming Languages: Python, Java, C, C++
  • Machine Learning & Computer Vision: Object Detection, Pose Estimation, YOLO, Explainable AI (XAI), PCA, NMF
  • Frameworks & Libraries: PyTorch, TensorFlow, scikit-learn, OpenCV, NumPy, Pandas
  • NLP & Generative AI: RAG, Transformers, Hugging Face, CamemBERT, Embeddings
  • Data Engineering & Cloud: AWS (EC2, S3), SQL, PostgreSQL, MongoDB, Apache Hop, Data Visualization (Matplotlib, Plotly)
  • Development Tools: Git, GitHub, Linux, Conda, Jupyter Notebook, Google Colab, VS Code
  • Web Technologies: HTML, CSS, JavaScript

Français

À propos

Ingénieur en Intelligence Artificielle et Machine Learning, avec une expérience pratique dans le développement, l'évaluation et l'analyse de modèles de machine learning et de deep learning. Je travaille actuellement chez Airbus sur l'IA explicable appliquée à la vision par ordinateur, avec également une expérience en NLP, data science et machine learning appliqué.

Expériences professionnelles

Stagiaire en Intelligence Artificielle — IA de confiance · Airbus SAS
Toulouse, France · Mars 2026 – Septembre 2026

  • Adaptation d'une méthode d'IA explicable à un modèle YOLO-NAS Pose multi-échelle sous PyTorch afin d'identifier les caractéristiques visuelles influençant les détections du modèle.
  • Développement d'un protocole expérimental pour extraire, sélectionner et analyser les représentations visuelles du modèle, en évaluant leur qualité, leur stabilité et leur capacité à préserver l'information utile aux détections.
  • Évaluation quantitative de la fidélité des détections à l'aide de métriques de détection et de correspondance spatiale, avec 90,5 % de précision et un PR-AUC de 0,776.
  • Analyse du comportement du modèle à différentes distances de la piste afin d'identifier les informations visuelles les plus influentes et leur évolution au cours de l'approche.

Stagiaire Data Scientist · Responsibio
Caen, France · Mai 2025 – Août 2025

  • Collecte et préparation de données audio de vocalisations de souris et de leurs métadonnées afin de construire un jeu de données pour la classification du bien-être animal.
  • Développement d'un pipeline de classification audio utilisant des mél-spectrogrammes, l'extraction de caractéristiques, une réduction de dimension par PCA et un SVM linéaire.
  • Évaluation du modèle avec une validation croisée à 5 folds, réduction de 2048 à 58 caractéristiques en conservant 86 % de la variance et obtention d'environ 79 % de prédictions correctes.

Projets principaux

PythonPyTorchYOLOv8OpenCVByteTrackHomographie

  • Création et annotation de jeux de données pour détecter les joueurs, gardiens, arbitres, ballon et points clés du terrain, puis entraînement de modèles YOLOv8 spécialisés.
  • Développement d'un pipeline de suivi vidéo combinant ByteTrack, filtrage et interpolation du ballon, avec projection des positions sur un terrain 2D par homographie.

PythonPyTorchHugging Face TransformersCamemBERTPandas

  • Développement d'un système NLP pour identifier automatiquement, mot par mot, les passages de mémoire épisodique et sémantique dans des récits cliniques en français.
  • Fine-tuning et comparaison de CamemBERT, ModernCamemBERT et DeBERTa, avec validation croisée par patient afin d'évaluer la généralisation du modèle sans fuite de données.

Autres projets

Compétences techniques

  • Langages de programmation : Python, Java, C, C++
  • Machine Learning & Computer Vision : Object Detection, Pose Estimation, YOLO, Explainable AI (XAI), PCA, NMF
  • Frameworks & bibliothèques : PyTorch, TensorFlow, scikit-learn, OpenCV, NumPy, Pandas
  • NLP & IA générative : RAG, Transformers, Hugging Face, CamemBERT, Embeddings
  • Data Engineering & Cloud : AWS (EC2, S3), SQL, PostgreSQL, MongoDB, Apache Hop, Data Visualization (Matplotlib, Plotly)
  • Outils de développement : Git, GitHub, Linux, Conda, Jupyter Notebook, Google Colab, VS Code
  • Technologies Web : HTML, CSS, JavaScript

Languages / Langues

  • Albanian / Albanais - Native / Langue maternelle
  • French / Français - C2
  • English / Anglais - C1

Resume / CV

Popular repositories Loading

  1. Web_projet Web_projetPublic

    Création d'une page web portfolio en utilisant HTML, CSS, JavaScript

    Python

  2. L2-ALGO-NUM L2-ALGO-NUMPublic

    C

  3. L2-SI L2-SIPublic

    TeX

  4. LOGIQUE-L2 LOGIQUE-L2Public

    Prolog

  5. L2-JEU-DE-SOKOBAN L2-JEU-DE-SOKOBANPublic

    C

  6. L2-POO L2-POOPublic

    Java