Senior AI Engineer with over 8 years of experience building and deploying advanced machine learning models from concept to production. I have a proven track record in pioneering novel algorithms and architectures in generative AI and multimodal systems, and I'm passionate about building transformative, cutting-edge AI that seamlessly integrates into real-world workflows.
π€ AtaraxAI: High-Performance, Fully-Offline AI Assistant
- Description: Architected and built a cross-platform, high-performance, fully-offline AI assistant from the ground up.
- Core Feature: Features a custom, low-latency C++ core for efficient inference on consumer hardware.
- Stack: Python, C++, llama.cpp, whisper.cpp, Tauri (React + Rust).
- Repo: https://github.com/zyannick/Atarax-AI/
π¬ LogosRL: Finetuning Mixture-of-Experts LLMs with Reinforcement Learning
- Description: A from-scratch framework to finetune MoE LLMs using RL algorithms (PPO, A2C, SAC) for mathematical reasoning and protein description generation.
- Key Innovation: Implements tracking of dead experts by monitoring usage during training.
- Repo: https://github.com/zyannick/LogosRL
π PathFormer-Reproduction: Time Series Forecasting on AWS
- Description: A complete reproduction of the PathFormer model (ICLR 2024) for time series forecasting, deployed on AWS infrastructure. This project showcases the full lifecycle from research reproduction to cloud deployment.
- Repo: https://github.com/zyannick/PathFormer-Reproduction
π§ Aurelius: A Deep Learning Framework from Scratch
- Description: A foundational Deep Learning framework built entirely from scratch to demonstrate core principles, heavily leveraging C++, Eigen, and SIMD for performance optimization.
- Repo: https://github.com/zyannick/aurelius
- Bioaster (Senior MLE):
- Architected a novel deep learning model, achieving state-of-the-art accuracy (+9%).
- de novo peptide design, identifying 100 therapeutic candidates for in-vitro testing.
- Genomic Vision (Computer Vision Engineer):
- Authored novel methods for image reconstruction, increasing Recall by 10 points and reducing data analysis time by 30%.
- Developed real-time C++/CUDA algorithms that enabled a 1.5x increase in sample throughput.
- Deployed high-throughput inference services with Docker & ONNX, reducing latency by 20%.
- Postdoctoral fellow in Machine Learning
- Authored early detection models for infection prediction from ECG signals (48 hours before infection detection).
- Interpreted ML models using explainability libraries (Captum) to ensure trust and interpretability.
- PhD in Computer vision:
- Designed efficient 3D reconstruction algorithms using stereo vision and SLAM principles on low-resolution thermal imagery.
- Proposed robust calibration methods and sub-pixel matching techniques improving reconstruction accuracy in challenging environments.
- Proposed robust calibration methods and sub-pixel matching techniques improving reconstruction accuracy in challenging environments.
- Ensta Paristech (Intern):
- Boosted real-time 3D SLAM pipelines with SIMD parallelization, achieving a 5x speedup vs. the OpenCV baseline.
- ICOMG (Software Engineer):
- Designed and built J2EE and ASP.NET applications (SQL, Javascript, C#, Java, Git, UML, Agile)