Deep Learning Engineer

Mars 2023 - Present

Internship @ Stellantis | Paris, France

• Reduced by 74% training time, while increasing by 11% the performance of state-of-the-art model DenseTNT.
• Improving an internally developed model PreTR, by adding multimodality, and allowing for a scene-aware backbone.
• Adapted state-of-the-art vehicles trajectory prediction models, Vecternet, DenseTNT, for pedestrian's trajectory prediction.
• Extracted map information of the Dataset ETH/UCY from raw images and converted it to a vector representation.

Computer Vision Engineer

Feb. 2023 - Mars 2023

Freelance @ Flaggr | Paris, France

• Reduced by 60% inference time of the main vehicle detection algorithm, while losing less than 4% accuracy.
• Automated the evaluation of new models by creating a Benchmark pipeline.
• Implemented their first version of a night vehicle detection algorithm.

Research Intern

May 2022 - Aug. 2022

Internship @ CNRS | Paris, France

• Designed a pipeline allowing real-time AI content generation for Augmented Reality, by integrating state-of-the-art text-toimage diffusion models and AR motion tracking techniques.
• Investigated and implemented state-of-the-art methods for diffusion-based image generation, upscaling, and segmentation.

Research Student

Nov. 2021 - Feb. 2022

TER @ Université Paris-Saclay | Paris, France

• Launched, under the supervision of Isabelle Guyon, the ML competition TRUSTAI on CodaLab.
• Prepared the Dataset used in the competition, building the preprocessing pipeline, and rigorously following strict guidelines.
• Explored and reviewed more than 10 datasets, diagnosing potential sources of discriminatory biases.
• Determined a bias evaluation metric, after inspecting the scientific literature on fairness in machine learning.

Software Engineer Intern

Internship @ LIRMM - Research Lab | Montpellier, France

• Led a team of 5 members as a Scrum master, following Agile software development practices and methodologies.
• Designed and built a cross-platform Vocal Assistant, using a microservices architecture with Python, Docker, and Flask.
• Delivered in less than 3 months a Raspberry Pi prototype, along with both a design and a technical documentation.

Full-Stack Developer Intern

June 2020 - July. 2020

Internship @ LIRMM - Research Lab | Montpellier, France

• Implemented a digital version of Labo DataViz, by coordinating a team of 3 to develop an interactive Web Platform.
• Collaborated with a team from the University of Quebec at Chicoutimi to establish the project specification.

App Developer Intern

June 2019

Internship @ Perform VR | Montpellier, France

• Optimized and tested a virtual reality android mobile application with Java.

Master - Artificial intelligence

2021 - 2023

Université Paris-Saclay Paris, France

Main Courses: ML Algorithms, Deep Leaning, Computer Vision, Large-Scale Distributed Data Processing, Probabilistic Generative Models, Applied statistics, Advanced Optimization, Signal Processing, NLP, Information Retrieval, Reinforcement Learning.

see transcript

Bachelor - Computer Science

2018 - 2021

University of Montpellier Montpellier, France

Main Courses: Algorithms & Data Structures, Project Management, System design, Object-oriented programming.

DELF B2 - French Language

2017 - 2018

Accent Français Montpellier, France

Academic Tutor

November 2022 - February 2023

University of Paris-Saclay Paris, France

Bias Mitigation For Age Detection

October 2022

Estimating the age from images while tacking the bias with respect to the protected attributes (Age, Gender, Ethnicity, Face Expression). Two main approaches were used in order to reduce the bias :

• Data Augmentation using GAN based models.
• Custom Loss using class Weight and MinDiff

• Language : Python
• Tools : TensorFlow, PyTorch, OpenCV, NumPy, plotly.
• GitHub : Bias Mitigation For Age Detection

Probabilistic Generative Models

October 2022

Implementing several probabilistic generative models from scratch using PyTorch. Three models were implemented :

• Variational AutoEncoder Notebook
• Restricted Boltzmann Machine Notebook
• Real-NVP Normalizing Flows Notebook

• Language : Python
• Tools : PyTorch, NumPy, Matplotlib
• GitHub : Probabilistic Generative Models

Signal Processing - Denoising

October 2022

Implementing several signal processing techniques for image and audio signal denoising. The implemented techniques are :

• Fourier Transform Notebook
• FIR IIR Filters Notebook
• Short-time Fourier Transform Notebook
• Spectral Density Estimation Notebook
• Wavelet Transform Notebook

• Language : Python
• Tools : Scipy, PyWavelets, NumPy, Matplotlib.
• GitHub : Signal Processing - Denoising

Twitter Streaming Data

May 2022

Processed, clustered, indexed, and queried tweets from a simulated data stream.

• Language : Python
• Tools : Spark, Elasticsearch, TwitterAPI, Socket.
• GitHub : Twitter Streaming Data

Hyponymy Extraction

April 2022

Named entity recognition (NER) and Detecting Hyponym\Hypernym relationship on the dataset of tech Patents.

• Language : Python
• Tools : Spacy, NLTK, Prodigy, Pandas, NumPy.
• GitHub : Hyponymy Extraction
• TowardsDataScience : Improving Named Entity Recognition (NER) | Implementing Hearst Patterns

Deep Network

April 2022

An implementation of a Deep Neural Network (MLP) without PyTorch. Implementing Backpropagation & momentum SGD optimizer.

• Language : Python
• Tools : NumPy, Matplotlib.
• GitHub : Deep Network<

Sentiment Analysis

February 2022

A sentiment analysis for Amazon food reviews using Bert, different embeddings and other NLP approaches. The goal was to classify reviews into three classes: Positive, Negative and Neutral. This project was done in a team of two in the context of a class on NLP during the master of AI at the university of Paris-Saclay.

• Language : Python
• Tools : PyTorch, Gensim, Transformers, Bert, NLTK, Scikit-Learn, NumPy, Pandas, plotly, Matplotlib, Seaborn, Imblearn.
• Difficulties & challenges :
  • Dealing with the imbalanced data, since the neutral class had very few examples.
• GitHub : Sentiment Analysis

Aerial challenge

December 2021

Aerial is a machine learning challenge on CodaLab Platform. It's a multi-class geographic image classification. I worked in a team of 3 to solve the challenge using Deep learning and we were able to achieve a score of 93% using a convolutional neural network.

• Language : Python
• Tools : TensorFlow, Scikit-Learn, NumPy, Pandas, Matplotlib
• GitHub : Aerial challenge

To-Be challenge

November 2021

To-Be is a machine learning challenge on CodaLab Platform that aims to adress the problems of medical imbalanced data classification. I worked in a team of 3 to solve the challenge and we were able to achieve a score of 77% using LGBMClassifier and we classed in top 3.

• Language : Python
• Tools : Scikit-Learn, NumPy, Pandas, Matplotlib, Seaborn, Imblearn
• Difficulties & challenges :
  • Dealing with the imbalanced data.
• GitHub : To-Be Challenge

Vocal assistant

January 2021 - August 2021

A desktop vocal assistant application. I was the scrum master of the team developing the application during the last year of my bachelor degree. I was also offered a two months paid internship during the summer to continue working on the project.

• Language : Python, JavaScript
• Tools : Docker, Tensorflow, NLTK, OpenCV, Flask, jQuery, Jira
• Difficulties & challenges :
  • Making the application compatible with Windows, Linux, MacOS.
  • Implementing the application on a Raspberry Pi.
  • Leading a team of 5 members.
• GitHub : Vocal Assistant

Labo DataViz

June 2020 - July 2020

An interactive web platform implementing a digital version of "labo DataViz" presented by UQAC (Université du Québec à Chicoutimi) in 2019. I was the scrum master of the team developing the platform during my internship in LIRMM (a research laboratory).

• Language : JavaScript, HTML, CSS, Python
• Tools : MongoDB, jQuery, Flask, anime.js
• Difficulties & challenges :
  • Making the platform Intuitive and easy to use.
  • Leading a team of 3 members.
• GitHub : Labo DataViz

Geoscape

January 2020 - April 2020

A web platform that scrapes the subreddit EarthPorn looking for images, tries to recognize the lication based on the caption, and then display the images on an interactive map. I was part of the team developing this platform during my bachelor degree.

• Language : Python, JavaScript, HTML, CSS
• Tools : MongoDB, jQuery, Flask, NLTK, Trello
• Difficulties & challenges :
  • Adapting to some changes in the Reddit API that occurred during the development.
  • Extracting the location of the images from the captions.
• GitHub : Geoscape

The EscaPysts

January 2019 - April 2019

A video game developed in Python and compatible with Windows, Linux and MacOS. I worked on this game in a team during the first year of my bachelor degree.

• Language : Python
• Tools : Pygame
• Difficulties & challenges :
  • Implementing an algorithm for finding the shortest paths.
  • Optimizing the game for faster level loading.
• GitHub : The EscaPysts