Luka Krsnik

Luka Krsnik

Software developer | ML Engineer

Download CV

About Me

Hi! I'm Luka, a curious engineer who wants to create something meaningful. This need led me to computer science, which is filled with puzzles you can solve with nothing but an empty notepad and a little ingenuity. I am also very intrigued, by the idea to not just create something from nothing, but something that creates something from nothing! These realizations fuel my passion for both computer and data science to this day.

Bio

Email
luka@outsmartify.com
Address
Ljubljana, Slovenia

Portfolio

Data Extraction / Natural Language Processing / Text Analysis / Tool Development

STARK - Dependency Parsing Statistics Extractor

A tool for dependency-tree extraction and analysis. It extracts and counts subtrees from given corpora using multi-core processing capabilities.

Data Extraction / Natural Language Processing / Text Analysis / Tool Development

CORDEX - a collocation extraction library

CORDEX is a Python library for extracting collocations based on user-defined rules.

Data Science / Machine Learning / Natural Language Processing / Text Analysis / Tool Development

Classla

Development of a fork of Stanza library that focuses on South Slavic languages. It enhances core functionalities such as tokenization, part-of-speech tagging, lemmatization, dependency parsing, and named entity recognition. I implemented some rule-based techniques, retrained models, and introduced semantic role labeling models.

Data Science / Machine Learning / Natural Language Processing / Research

Enhancing deep neural networks with morphological information

Explored the enhancement of deep neural network models by adding morphological features for downstream tasks. These included named entity recognition, dependency parsing, and comment filtering across multiple languages. Notable performance improvements were achieved in LSTM-based models, while BERT-based models showed limited enhancement.

Data Science / Machine Learning / Natural Language Processing

Standardness

This project focused on identifying nonstandard language in Slovenian tweets. We used SVMs and BERT-based neural networks, to predicted a unique standardness metric, that was used for selection of tweets for manual annotations.

Natural Language Processing / Research

Cross-Lingual Embedding Alignment Using Anchor Points

This research project explored cross-lingual embedding alignment using anchor points.

Data Extraction / Text Analysis / Tool Development

LIST - Corpus analyzer

Development of the LIST corpus extraction tool, a program designed for extracting statistics from text corpora. It offers summarizations at various linguistic levels and is actively used for comparisons between different corpora.

Data Science / Machine Learning / Natural Language Processing / Research

Stress assignment

This project tackled the task of stress assignment for Slovene words, utilizing machine learning and deep neural networks to achieve an accuracy of 87.62% for stress predictions. The model was used for annotations in the largest morphological lexicon for Slovenian, enhancing accessibility and accuracy in stress assignment for linguists and language enthusiasts.

Contact

luka@outsmartify.com