# Aleksandar Tomasevic

> Data Scientist and Applied AI Researcher based in Novi Sad, Serbia. Current roles: Senior Research Associate, Scientific Computing Laboratory at Institute of Physics Belgrade; Assistant Professor / Research Associate at University of Novi Sad. Primary topics: Online-community analysis, Machine learning, Natural language processing, Text classification, Sentiment analysis, Toxicity analysis.

## Personal

- Canonical name: Aleksandar Tomašević
- Alternate spelling: Aleksandar Tomasevic
- Email: atomashevic@gmail.com
- Phone: (+381) 69 1196392
- Website: https://atomasevic.com
- Titles: Data Scientist, Applied AI Researcher
- Location: Novi Sad, Serbia

## Work History

### Senior Research Associate, Scientific Computing Laboratory — Institute of Physics Belgrade

- Date range: 2025-07-01 to present
- Location: Belgrade / Serbia
- Organization URL: https://www.ipb.ac.rs/
- Built computational workflows for large-scale data analysis, simulation, and model evaluation in Python and R.
- Developed and validated machine learning, NLP, and network-analysis pipelines for behavioral and online-community data.
- Worked on AI-system evaluation problems involving trust, toxicity, conversation quality, agent behavior, and online-community dynamics.
- Contributed to technically complex platforms using PostgreSQL, Dockerized services, Linux, and reproducible experiment workflows.

### Assistant Professor / Research Associate — University of Novi Sad

- Date range: 2016-02-01 to present
- Location: Novi Sad / Serbia
- Organization URL: https://www.uns.ac.rs/
- Led data-intensive projects using Python and R to analyze large behavioral, textual, and network datasets.
- Designed machine learning and NLP workflows for online-community research, including text classification, sentiment analysis, toxicity-related analysis, and simulation support.
- Built automation scripts and reproducible pipelines for data cleaning, modeling, visualization, reporting, and experimental analysis.
- Collaborated with physicists, psychologists, and sociologists on technically demanding computational research.

### Fulbright Visiting Researcher — University of Virginia, School of Data Science

- Date range: 2023-10-15 to 2024-01-31
- Location: Charlottesville, VA / USA
- Organization URL: https://www.virginia.edu/
- Developed `transforEmotion`, an R package with Python backend for multimodal emotion recognition across text, image, and video data.
- Built reproducible machine learning workflows connecting transformer-based models, statistical analysis, and large-scale media data processing.

### Medior Game Designer — Eipix Entertainment

- Date range: 2014-05-01 to 2016-02-01
- Location: Novi Sad, Serbia
- Organization URL: https://www.eipix.com/
- Worked with cross-functional production teams on game systems, content design, and delivery across 10+ shipped titles.

## Education

### PhD in Quantitative Methods — University of Novi Sad

- Date range: 2012-10-01 to 2019-03-08
- Location: Novi Sad, Serbia
- Institution URL: https://www.uns.ac.rs/
- Research and training centered on statistics, computational methods, network analysis, and large-scale behavioral data in social sciences.

### Master in Social Sciences — University of Novi Sad

- Date range: 2011-10-01 to 2012-10-01
- Location: Novi Sad, Serbia
- Institution URL: https://www.uns.ac.rs/

### Bachelor in Social Sciences — University of Novi Sad

- Date range: 2007-10-01 to 2011-06-30
- Location: Novi Sad, Serbia
- Institution URL: https://www.uns.ac.rs/

## Projects

### YSocial Multi-Agent Social Platform Simulation

- Affiliation: Institute of Physics Belgrade / Research collaboration
- Date range: 2025-09-01 to present
- Built and evaluated components of an LLM-powered social media twin spanning microblogging and forum-style simulations for trust, toxicity, and conversation-quality research.
- Worked on agent memory, retrieval, and behavioral parameterization to improve reply selection, interaction pacing, and coherence in multi-agent threads.
- Contributed to a Python/Flask/SQLAlchemy stack with SQLite and PostgreSQL backends, covering experiment metadata, agent populations, recommendations, posts, reactions, and agent-opinion tracking.
- Used Dockerized PostgreSQL services and OpenAI-compatible backends including Ollama and vLLM to support simulation runs, annotation workflows, and system evaluation.
- Investigated concrete failure modes such as repetitive outputs, thread-structure breaks, and parent-child mismatches, then used real community data to guide validation and calibration.

### MADOC Dataset and Tools

- Affiliation: Institute of Physics Belgrade
- Date range: 2025-01-01 to present
- URL: https://zenodo.org/records/14637314
- Built a research dataset and tooling stack for large-scale online-community analysis, with emphasis on reproducible ingestion, cleaning, and downstream modeling workflows.
- Implemented NLP pipelines for text classification, toxicity detection, and sentiment analysis over cross-platform community text.
- Developed `pyMADOC`, a Python package for dataset access and text processing using pandas, NumPy, and scikit-learn.

### transforEmotion R package

- Affiliation: Personal collaboration
- Date range: 2023-10-01 to present
- URL: https://github.com/atomashevic/transforEmotion
- Built an R package with a Python backend for multimodal emotion recognition across text, image, and video, with more than 9k CRAN installs.
- Integrated transformer-based models into reproducible data-processing workflows for large-scale media analysis.
- Maintained the package as production-quality open-source software with GitHub Actions CI/CD and cross-language tooling.

### COVID-19 Data Analysis and Visualization Platform

- Affiliation: Data Versus Corona initiative
- Date range: 2020-04-01 to 2020-06-30
- URL: https://scienceversuscorona.shinyapps.io/covid-overview/
- Engineered R data pipelines to clean, analyze, and visualize epidemiological data from multiple sources.
- Built a real-time data fetching and processing API for automated updates and reproducible reporting.
- Created interactive visualizations for communicating complex health data clearly to broad audiences.

## Skills

### Data Science & Machine Learning

- Python: pandas, NumPy, scikit-learn, transformers, PyTorch, Flask, SQLAlchemy
- Machine Learning: feature engineering, fine-tuning, benchmarking, evaluation
- NLP: text classification, sentiment analysis, embeddings, topic extraction, toxicity detection
- Data Visualization: Plotly, matplotlib
- R: statistical analysis, package development
- SQL: PostgreSQL, SQLite, MySQL

### Cloud & Tools

- Linux and CLI-first workflows
- Scientific Computing: HPC environments, remote compute, reproducible pipelines
- Ops: Docker, GitHub Actions CI/CD
- LLM Systems: Ollama, vLLM, OpenAI-compatible APIs
- Version Control: Git and GitHub workflows
- Development Environments: JupyterLab, VSCode, Neovim

### Research & Communication

- Experimental design and validation
- Communication in interdisciplinary technical teams
- Technical writing and documentation
- Problem-solving and analytical thinking

## Languages

- English: C1
- French: B1
- Serbian: Native

## Interests

- Linux geek
- Football statistics and analytics: xG analysis, passing networks, player tracking data
- Video games: console (PS4, Switch), retro gaming
- Board games: Everdell, The Isle of Cats, Cartographers

## Profiles

- [GitHub (atomashevic)](https://github.com/atomashevic)
- [LinkedIn (atomasevic)](https://www.linkedin.com/in/atomasevic/)
- [X (atomasevic)](https://twitter.com/atomasevic)

## Machine-Readable Data

- [Source YAML](https://atomasevic.com/cv.yml)
- [Structured CV JSON](https://atomasevic.com/cv.json)
- [Person JSON-LD](https://atomasevic.com/person.jsonld)
- [Short LLM profile](https://atomasevic.com/llms.txt)