May 2020 – Jan 2021 · 8 mos · New York, United States · On-site

• At Proxima, a pioneering AI drug discovery company backed by NVIDIA Healthcare I:
• Developed advanced AI-driven drug discovery AI models through innovative machine learning solutions.
• Developed predictive models to identify potential drug candidates by analyzing biochemical and molecular datasets.
• Designed and trained neural networks for molecular property prediction, optimizing deep learning architectures for high accuracy.
• Implemented data preprocessing pipelines and feature engineering strategies tailored to complex pharmaceutical data.
• Collaborated closely with cross-functional teams of biologists and chemists to validate and refine model outputs for real-world application.

Oct 2017 – Feb 2020 · 2 yrs 4 mos · San Francisco Bay Area

• 🧠 Co-authored Mi-Prometheus, IBM’s open-source deep learning framework based on PyTorch, published at NeurIPS 2018.
• → Publication link: https://openreview.net/forum?id=rkxVOvH3FQ
• 📈 Designed and deployed ML solutions for CANON and HITACHI Metals:
• → Built full industrial data pipelines, trained and optimized CV models, and deployed them to production.
• 🤖 Led Visual Question Answering (VQA) research on CLEVR dataset:
• → Developed a CNN-LSTM model achieving 99% accuracy, with successful transfer learning on CLEVR-CoGenT.
• → Published at NeurIPS 2018: https://arxiv.org/abs/1811.06529
• 🎥 Pioneered Video Question Answering (Video QA) research on COG dataset:
• → Built a novel deep learning model, achieving 97.5% accuracy.
• → Published at NeurIPS 2019, Vancouver.
• 🔬 Google Scholar Profile: https://scholar.google.com/citations?user=xe6e3pAAAAAJ&hl=en
• Technical Skills
• Programming: Python, C, C++, Matlab, JavaScript, Bash, Objective-C, Assembly
• ML Frameworks: PyTorch, TensorFlow, Keras, OpenCV, Scikit-Learn, Pandas
• Tools: LaTeX, Overleaf, GnuPlot, Maple, ANSYS, Abaqus, Vivado FPGA, Arduino
• Web Dev: React, PHP | Version Control: Git, GitHub, Bash
• Expertise
• Object Detection & Recognition (RGB, 3D RGB-D, Video Data)
• Visual and Video Question Answering (VQA/Video QA)
• Deep Neural Networks, Reinforcement Learning
• Models: ResNet, BERT, Transformer, XLNet, Transformer-XL, Inception V3
• Highlights
• 📚 2 NeurIPS publications (2018, 2019)
• 🏆 Built production-grade ML systems for global industrial clients
• 🔬 Advanced multimodal deep learning research (NLP + Computer Vision)

Jan 2017 – Oct 2017 · 9 mos · San Francisco Bay Area

• 🔬 Hyperimager Project – Collaborative Initiative with IBM New York Hyperimager Team
• Led the development of a cutting-edge multi-modal imaging device integrating RGB, infrared, and radar sensors, culminating in the patented innovation “Method for Guided Multi-Spectral Inspection” (📄 Invention Ref: P201810340, Record ID #95272906).
• 👨‍🔬 Co-inventors: Alberto Valdes Garcia, Ahmet Serkan Ozcan, Vincent Albouy, Asaf Tzadok, Petar K. Pepeljugoski, Jean-Olivier Plouchart.
• Key technical achievements:
• 🤖 Designed and deployed an object detection algorithm using RGB and RGBD data with TensorFlow.
• 🧠 Built models for action recognition, part segmentation, and semantic segmentation using RGB input.
• 🔁 Developing a cross-modal attention network to improve multi-spectral object detection.
• ⚙️ Implemented hardware acceleration on FPGAs using VHDL and High-Level Synthesis (HLS) for real-time processing.
• 🤝 Machine Learning Vision Systems – Industry-Funded Project (CANON, Samsung, Honda, BMW, HTC)
• Spearheaded a high-impact machine learning initiative co-funded by leading imaging and automotive companies.
• Key contributions:
• 👨‍💼 Led a team of 4 developers to engineer a vision-based ML solution for CANON.
• 🎯 Created an attention mechanism algorithm that intelligently prioritizes relevant features in image data, enhancing efficiency and precision.

Mar 2016 – Dec 2016 · 9 mos · San Francisco Bay Area

• Artificial Intelligence and Machine Learning project funded by the American government (DARPA)
• Studied IBM's Context Aware Learning (CAL) and Numenta's HTM
• Implemented a machine learning algorithm based on CAL to enable a humanoid robot to mitigate is loss of balance when such an anomaly was detected (C++, Python)
• Designed a robot named KATE and implemented a simulation of KATE using ODE Library
• Led a 3 developers team to build three KATE robots (5 feet tall) and embed our Machine Learning solution
• As a result, we improved the robot's meantime to failure by a factor of 10

Nov 2015 – Apr 2016 · 5 mos · Région de Lyon, France

• Developed a high performance thermal insulator to save 50% on heating and air-conditioning energy
• Programmed a simulation using finite elements method on ANSYS
• Used modeling software: Abaqus, SolidWorks
• Characterized thermal, mechanical and electrical behaviors in laboratory

Jan 2015 – Sep 2015 · 8 mos · Région de Valence, Espagne

• Managed sport and music professional events averaging 500 attendees
• Planned the marketing, sales and communication strategy - Video editing (Software : Final Cut)

Aug 2011 – Jun 2012 · 10 mos · Région de Lyon, France

• Managed INSA Lyon's shop and sales (+$30k turnover)
• Organized a fashion show for INSA Lyon's prom gala 2013 (4000 attendees)

Jan 2010 – Jun 2014 · 4 yrs 5 mos · France, South-America, North-America

• International Competitor (Alpine Skiing)
• Several university European-cup podiums
• Third place at France university championship

Jan 2008 – Oct 2014 · 6 yrs 9 mos · Val Thorens

• Instructed ski and snowboard lessons at Val Thorens Club Med
• Supervised groups and managed ski events from management to marketing strategy