I bring State of the Art Machine Learning models from PoC to production.
ML, NLP, OpenAI, Pytorch, Tensorflow, ONNX, Hugging Face, Spark ML
Skypilot, DVC, BentoML, ClearML, Mlflow
Docker, Gitlab CI, Kubernetes, Helm, Argo CD, Terraform
4 years experience in building projects on AWS.
Python, Django, Celery, Fastapi
Streamlit, Grafana, Kibana, Tableau
Dagster, Airflow, Hadoop, Spark, Snowflake, Elastic Search
I'm a Machine Learning Engineer with a strong mathematical background. Along the way, I realized that making ML models usable in products is what stimulates me.
I worked at Ubisoft for three years deploying Machine Learning models to detect in-game fraudulent transactions. This experience taught me how to manage end-to-end Machine Learning projects, with a in-depth exploration of Infrastructure, Data Engineering and MLOps matters.
Then, I joined GitGuardian as a Machine Learning Engineer, where I helped to improve the Secrets Detection Engine. At the boostrap of the ML team, I built the MLOps stack of the company from scratch. Then, I worked on deploying fine-tuned Large Language Models in AWS EKS and integrating these deployments in the company main product, the private git repositories secret detection tool. I also did PoCs to automatically replace hardcoded secrets in code with environment variables adapted to various programming languages, involving OpenAI API and the AST used by the Black formatter.
I am now starting my journey as a Freelance Machine Learning Engineer. If you want to have a talk, please contact me on Linkedin or at michael.romagne@gmail.com.
- Built the MLOps stack from scratch: Gitlab CI, SkyPilot, DVC pipelines, Dagster jobs, Streamlit, ONNX Runtime, BentoML, Helm, ArgoCD.
- Improved the Secrets Detection Engine by fine-tuning LLMs and deploying them on AWS. I integrated these models in the main product of the company, the Secret Detection tool, dividing the number of errors made by the detection engine by 5.
- Developed PoCs on automatic remediation for leaked secrets (OpenAI API, Black AST and code formatting).
- End-to-end Fraud Detection project in e-commerce transactions (Ubi Connect and Steam).
Led Research tasks (Feature Engineering, Semi-supervised learning) and put in place MLOps best practices (DVC, remote jobs on K8s, ClearML, model inference on AWS).
The project led to 5% of net sales savings, about 4 millions euros per year, compared to the previous fraud detection product.
- Time Series forecasting on Acquisition, Retention, Monetization and Ubisoft servers vCPU usage. Trained and deployed Generalized Additive Models to improve forecasts.
Research on Digital Twins to optimize IoT Systems. Data Science, Simulation and Monitoring of IoT systems.
NLP on Orange mobile phone and internet boxes logs to predict churn and customer satisfaction.
