Peter is an experienced Software Developer and Researcher, providing hands-on backend development experience in Python to deploy packages and actively participating in AI, NLP and CV research. He contributes to software design, architecture, development, maintenance and production support - creating strategies to identify and mitigate risks, resolve defects and build highly scalable distributed systems for clients.
Developed a Machine Learning NLP-based model for extreme multi-label email classification with a web interface for full management. Added researcher-less training and deployment where clients can train their models with their data through the web interface. The solution performs automatic EDA on the data, trains models created in PyTorch, and checks for their performances. Built the system using Python.
Built a new recommender platform that allows users to create their labels and assign them to any posts and comments on social media. It suggests to users the most probable label they should assign to a selected post - saving time and making the whole process easier. Worked on the environment for local development with Docker and Docker-Compose.
Worked on the machine learning processes in developing a siamese neural network and XGBoost models for the product pairing system. Improved the models, managed the system's architecture, explored Kafka for communication between its different parts, and deployed MLflow in Kubernetes for production experiment tracking. Ensembled models based on FastText and boosting trees, training pipeline, automatic data analysis for re-training, and deployment in production in Kubernetes using Docker and Helm with REST/RESTful APIs. Adopted standard DevOps principles on the project.
Education
PhD in Artificial Intelligence
Faculty of Electrical Engineering, Czech Technical University in Prague
2021 - Present (4 years)
MSc. Artificial Intelligence
Faculty of Electrical Engineering, Czech Technical University in Prague
