CV
Work Experience
Oct 2023. - Present
NLP Engineer at Newfire Global Partners, Zagreb, Croatia
- Interviewed and evaluated ML/NLP engineer candidates
- Trained, evaluated, and integrated Transformer-based multi-task relation linking model that reduced the false-positive extractions by 90%
- Trained, evaluated, and integrated Transformer-based multi-task relation linking model that 1) proved the usefulness of a multi-task approach for our service, 2) set the training and evaluation standards and guidelines for future multi-task models, and 3) provided foundation for currently 3 multi-task models in the service
- Trained, evluated and integrated a general, Transformer-based word sense disambiguation (WSD) language model that: 1) achieved 80% zero-shot accuracy, 2) replaced five specialized WSD models, and 3) improved the overall service
- Created CLIs for automatic data cleaning (removing duplicates, near-duplicates, and outliers) and consistency verification
Oct 2021. - Sep 2023.
Research Associate at Text Analysis and Knowledge Engineering Lab, University of Zagreb, Faculty of Electrical Engineering and Computing, Croatia
- Developed and deployed REST API for event extraction that processed over a million news articles over a 6 month period
- Trained, evaluated, and adapted Transformer-based language models for efficient and robust text classification and extractive question answering
- Improved the event extraction pipeline’s throughput by over 300% while maintaining extraction performance and robustness
- Lead and supervised two annotation projects, a text classification and named entity linking, with a team of 6 annotators in each project. Additionally, co-lead an annotation project (sequence labeling) with a team of 8 annotators
- Co-mentored 3 student’s BSc thesis and 2 student’s MSc thesis
- Teaching assistant for Artificial Intelligence course and Machine Learning course
Feb 2018. - Jun 2019.
Junior Robotics Software Engineer at Gideon (Brothers), Zagreb, Croatia
- Developed a centralized system for coordinating a heterogeneous fleet of autonomous robots
- Completed a comprehensive review of optical flow algorithms and models, and evaluated their potential for use on autonomous mobile robot
Summer 2017.
Game AI Developer Intern at Lion game Lion, Zagreb, Croatia
- Developed an AI, based on Monte Carlo Tree Search, for a turn-based strategy game that surpassed human performance.
Publications
- F. K. Došilović, M. Brčić and N. Hlupić, "Explainable artificial intelligence: A survey," 2018 41st International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO), 2018, pp. 0210-0215, doi: 10.23919/MIPRO.2018.8400040.
Education
University of Zagreb, Faculty of Electrical Engineering and Computing, Croatia
Sep 2019. – Feb 2022.
Master's in Computer Science
Thesis: Extraction of Articles from News Portals Using Machine Learning
Thesis: Extraction of Articles from News Portals Using Machine Learning
Oct 2015. – Sep 2019.
Bachelor's in Computing
Thesis: Object Tracking Using Camera Mounted on Mobile Robot
Thesis: Object Tracking Using Camera Mounted on Mobile Robot
Selected Projects
Personal Projects
Mar 2023.
gpustatus - a simple CLI for collecting info about GPU
A simple CLI, written in Go, for collecting info on GPUs from servers listed in .ssh/config.
A simple CLI, written in Go, for collecting info on GPUs from servers listed in .ssh/config.
Feb 2023.
jsonfmt - a simple CLI formatting large JSON files
A simple CLI, written in Nim, for formatting large JSON files.
A simple CLI, written in Nim, for formatting large JSON files.
Feb 2017. – May 2017.
molecule - a simple chess engine written in modern C++
A simple chess engine written in modern C++. Engine features standard components, such as: a) move generation module, based on magic bitboards, b) alpha–beta pruning and quiescence search, c) Zobrist hashing for detecting identical states during search.
A simple chess engine written in modern C++. Engine features standard components, such as: a) move generation module, based on magic bitboards, b) alpha–beta pruning and quiescence search, c) Zobrist hashing for detecting identical states during search.
University Projects
Fall 2020.
Merging statistically similar regions
Contemporary approaches for semantic segmentation use deep neural networks and achieve amazing results. Instead of using DNNs for segmenenting images, in this seminar I explore and expand on an algorithm for merging statistically similar regions, usually applied on grayscale images, applied on RGB images.
Contemporary approaches for semantic segmentation use deep neural networks and achieve amazing results. Instead of using DNNs for segmenenting images, in this seminar I explore and expand on an algorithm for merging statistically similar regions, usually applied on grayscale images, applied on RGB images.