Athanasios Tonios
Computer Science and Engineering graduate focused on data science, machine learning, and NLP.
About
Information now plays a central role in how the world moves forward, and I find working with it fascinating. What interests me most is how much we can retrieve from data, and what new possibilities that information opens up. That is why I enjoy analyzing data and designing solutions around what I find.
Looking ahead, I want to go deeper into NLP, language models, information retrieval, and graph analysis. My diploma thesis on fair community detection let me study how connected data can be analyzed while treating different groups fairly. I am especially interested in how retrieval and language models can work together to make large collections of text easier to use.
Projects
-
Fair Community Detection in Graphs
Developed Fair Louvain, an adaptation of the Louvain algorithm that finds well-connected communities while ensuring that groups defined by sensitive attributes, such as gender, are fairly represented in each one. Evaluated it against existing methods on synthetic graphs and real-world networks, measuring both community quality and fairness.
An illustration of my thesis topic, with two groups of nodes (and ). Balance is the smaller group's share against the larger one in a community, and 1.00 means equal representation. -
CNN News RAG Assistant
Local RAG assistant that answers questions over about 4,000 CNN articles (55,700 text chunks). Retrieves the five most relevant chunks with Sentence-Transformers embeddings and a FAISS index, generates grounded answers with a local LLM through Ollama, and shows the supporting text in a desktop GUI.
View on GitHub -
Articles Search Engine
Implemented a desktop search engine for CNN articles with Java and Apache Lucene, offering multi-field search, BM25 ranking, pagination, and query-term highlighting. Added personalized re-ranking that boosts results using terms from the user's recent searches.
View on GitHub -
Citation Prediction
Link prediction model that predicts citations between papers, using an XGBoost classifier on graph features, shared authors, and the semantic similarity of paper abstracts. Reached a log loss of 0.1149 on the Kaggle private leaderboard.
View on GitHub -
Greek++ Compiler
Python compiler that translates greek++, a small language with Greek keywords, into RISC-V assembly. Includes lexical analysis, a recursive-descent parser, semantic checks, a symbol table, and intermediate code generation.
View on GitHub -
SocialBookstore
Web application where users share books, with sign-up and login, book offers and requests, search, and personalized recommendations. Developed with Spring Boot and MySQL, with 40 unit tests.
View on GitHub
Skills and education
- Programming
- Python, Java, C, SQL
- ML and data
- PyTorch, pandas, NumPy, NetworkX, FAISS, Sentence-Transformers
- Search and frameworks
- Apache Lucene, Spring Boot, OR-Tools
- Areas
- Machine learning, NLP, information retrieval, data mining, graph and network analysis
- Education
- Diploma of Engineering (Integrated Master, 300 ECTS), Computer Science and Engineering, University of Ioannina, 2026
- Languages
- Greek (native), English (C2, ECPE)
CV
My education, projects, and skills on one page.