Curriculum Vitae
Filip Mirković
Machine Learning Engineer & Physicist
| Zagreb, Croatia | filip.mirkovic2211@gmail.com | GitHub |
Summary
Machine Learning Engineer experienced in research, development and deployment of production-grade deep learning systems. Specialized in audio signal processing, generative diffusion models, MLOps infrastructure, and high performance computing. Proven track record of successfully building solutions with high degrees of efficiency in resource scarce environments. Strong foundation in physics (MSc) and mathematics with hands-on experience in PyTorch, CUDA optimization, and end-to-end ML pipelines. Selected for a competitive machine learning program hosted by Google DeepMind (ML2School 2025), and held poster presentations.
Technical Skills
Machine Learning: Deep Learning, Neural Networks, CNNs, Transformers, GANs, VAEs, GNNs, Flow Matching, Random Forests, SVM, Ensemble Methods, Model Compression, Quantization
Programming: Python, PyTorch, PyTorch Lightning, TensorFlow, Scikit-learn, C++, CUDA, FastAPI, SQL
Data & Computation: NumPy, Pandas, SciPy, Matplotlib, Distributed Training, GPU Optimization, High-Performance Computing
MLOps & Infrastructure: Docker, MLflow, Git, CI/CD, Triton Inference Server, Linux, Hydra
Domain Expertise: Audio Signal Processing, Time Series Analysis, Computer Vision, Computational Physics, Statistical Modeling
Professional Experience
Machine Learning Engineer
Atomic Intelligence | Zagreb, Croatia | September 2023 - Present
- Designed and trained deep learning models for state-of-the-art Audio Source Separation, achieving 40% model size reduction while maintaining SOTA performance metrics compared to previous instances.
- Designed and implemented end-to-end ML training infrastructure using PyTorch Lightning, MLflow, and Hydra, reducing experiment iteration time by 50%.
- Led research planning for large-scale initiatives, defining technical roadmaps and deliverables aligned with EU funding requirements, securing €1.3M in funding.
- Built synthetic data generation pipeline for medical peptidomics research, enabling client to augment limited patient datasets while maintaining statistical fidelity and privacy compliance.
Student Teaching Assistant
University of Zagreb, Faculty of Electrical Engineering and Computing | February 2023 - June 2023
- Mentored 100+ undergraduate students in deep learning fundamentals, providing detailed feedback on lab assignments and projects.
Summer Student Intern
CERN | Geneva, Switzerland | June 2022 - September 2022
- Conducted comprehensive analysis of background effects in the BCM1F luminosity detector of the CMS experiment, identifying 3 primary background contributors.
- Performed statistical analysis of afterglow effects across the CMS detector array, processing terabytes of collision data.
Education
Master of Science in Physics
University of Zagreb, Faculty of Science | October 2018 - February 2024 | GPA: 4.6/5.0
Thesis: Applications of Deep Learning Methods in Complex Systems - Developed graph neural network frameworks for modeling complex systems with symmetry-aware architectures.
Relevant Coursework: Statistical Mechanics, Quantum Physics, Differential Geometry, Numerical Methods, General Relativity
Bachelor’s Program in Mechanical Engineering
University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture | October 2016 - Jul 2018 | Incomplete
Additional Coursework in Machine Learning
University of Zagreb, Faculty of Electrical Engineering and Computing | 2022 - 2023
Completed: Machine Learning 1, Deep Learning 1, Deep Learning 2
Projects
Audio VAE for Music Source Separation
Developed variational autoencoder architecture for music source separation using the GTZAN dataset. Achieved high-quality audio reconstruction with a compressed (×512) latent space suitable for a variety of downstream tasks.
Flow Matching for Generative Modeling
Implemented flow matching algorithm from scratch in PyTorch for continuous normalizing flows, demonstrating state-of-the-art generative modeling techniques.
End-to-End ML Pipeline Template
Built production-ready medical image segmentation pipeline featuring MLflow experiment tracking, Streamlit inference app, and containerized deployment with Docker.
Lumen Data Science Competition - 2nd Place
Developed a high-accuracy audio classification pipeline that placed 2nd at a competition hosting 50+ teams.
Languages
| Croatian (Native) | English (Fluent) |