Dario Stojanovski
Verified Expert in Engineering
Natural Language Processing (NLP) Developer
Skopje, Macedonia
Toptal member since October 28, 2021
Dario is a recently graduated PhD in computer science at LMU Munich, where he worked on neural machine translation and natural language processing. He has seven years of experience in NLP and ML contributing to various international research projects while working at an ML company and performing his internship at Amazon Research. Dario is interested in artificial intelligence and machine learning projects with a particular focus on natural language processing, as this is the field he is keen on.
Portfolio
Experience
- Machine Learning - 7 years
- Natural Language Processing (NLP) - 7 years
- Generative Pre-trained Transformers (GPT) - 7 years
- Python - 7 years
- Deep Learning - 7 years
- Neural Machine Translation - 4 years
- Text Classification - 3 years
- PyTorch - 3 years
Preferred Environment
Ubuntu, MacOS, Visual Studio Code (VS Code), Git
The most amazing...
...project I've developed is a context-aware machine translation model, competitive with the production models during my internship time at Amazon.
Work Experience
Senior Machine Learning Engineer
Loka
- Worked on developing proof-of-concept machine learning projects in NLP, computer vision, and speech processing.
- Contributed to machine learning projects for clients in various industries.
- Involved in writing project proposals and closely interacted with clients before and during project development.
NLP Researcher
Ss. Cyril and Methodius University in Skopje
- Worked on multilingual NLP models for identifying social media posts indicating an ongoing wildfire.
- Involved in researching and developing NLP models, data annotation, and deployment.
- Successfully achieved predefined goals for model performance across all languages despite a significant lack of annotated data.
Research Associate
LMU Munich
- Worked on neural machine translation under the supervision of Prof. Dr. Alexander Fraser.
- Contributed to context-aware and document-level NMT models to facilitate the integration of contextual information. Published papers on coreference resolution and coherence in context-aware NMT as well as its usefulness for domain adaptation.
- Implemented state-of-the-art NMT models in Nematus, Sockeye, and Fairseq.
- Worked on unsupervised NMT models. Participated in the last three WMT shared tasks on unsupervised translation and developed the best models in 2020.
- Co-authored papers on efficient language model pretraining and fine-tuning for UNMT.
- Supervised students working on their bachelor’s and master’s theses. Gave lectures in various classes and supervised students in several seminars.
Applied Research Intern
Amazon.com
- Worked at Amazon Alexa on a research project about context-aware NMT for Amazon Translate.
- Developed various context-aware NMT models and conducted a range of experiments and thorough evaluations.
- Wrote project proposals and project reports detailing the data preprocessing, model development, and evaluation.
Software Engineer
3PDevelopment
- Designed and implemented an API to facilitate the working of a mobile and a web application.
- Implemented an accompanying web application according to specified designs.
- Deployed the entire system in a production environment.
Junior Researcher
Ss. Cyril and Methodius University in Skopje
- Worked as a junior researcher on the FP7 project MAESTRA.
- Conducted research on sentiment analysis using deep learning techniques. Participated in the SemEval 2016 shared task on Twitter sentiment analysis with strong results across metrics.
- Developed several applications using the proposed DL methods: sentiment analysis in news-related tweets and geo-graphical social hotspots and emotion analysis in sport-related and local government-related tweets.
Experience
PhD Thesis on Context-aware Neural Machine Translation
The research proposes to improve state-of-the-art NMT models by including document-level information. The PhD thesis proposes several novel context-aware NMT models, how to better evaluate them, how to apply them for domain adaptation, and how to improve their training using curriculum learning inspired by how humans learn.
Unsupervised Neural Machine Translation
https://arxiv.org/abs/2010.13192#The project was a team effort that resulted in our model obtaining, by far, the best scores in the competition. The proposed system was based on state-of-the-art methods for unsupervised NMT, such as MASS pretraining and RE-LM fine-tuning, a state-of-the-art approach proposed by our team.
Education
PhD in Computer Science
LMU Munich - Munich
Master's Degree in Software Engineering
Ss. Cyril and Methodius University - Skopje, Macedonia
Bachelor's Degree in Informatics and Computer Engineering
Ss. Cyril and Methodius University - Skopje, Macedonia
Certifications
Machine Learning
Coursera
Skills
Libraries/APIs
PyTorch, Hugging Face Transformers, Keras, Theano, React
Tools
PyCharm, Git
Languages
Python, SQL, JavaScript, C#, HTML, CSS, Java
Frameworks
MXNet, ASP.NET MVC, AngularJS
Storage
Databases, MongoDB, MySQL, Elasticsearch
Platforms
Software Design Patterns, Ubuntu, MacOS, Visual Studio Code (VS Code), Amazon Web Services (AWS), Google Cloud Platform (GCP), Weights & Biases
Other
Machine Translation, Natural Language Processing (NLP), Transformers, Software, BERT, XLM-R, Excel Macros, Neural Machine Translation, Deep Learning, Text Classification, Sentiment Analysis, Artificial Intelligence (AI), Neural Networks, Generative Pre-trained Transformers (GPT), Recurrent Neural Networks (RNNs), Long Short-term Memory (LSTM), Fairseq, Algorithms, Data Structures, Nematus, Unsupervised Learning, Convolutional Neural Networks (CNNs), Google Earth, Machine Learning, Amazon Translate
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring