NLP and text-data engineer (Python). BA in applied/computational linguistics, Taras Shevchenko National University of Kyiv. I build text-processing tools end to end: data collection → cleaning → corpus building → retrieval and evaluation. Working languages: Ukrainian and English.
| Project | What it shows |
|---|---|
| lyrics-semantic-search | Hybrid retrieval (dense embeddings + BM25, fused by Reciprocal Rank Fusion) over a ~30k-song corpus. Cached offline build pipeline, pluggable embedders and index backends, 176 tests, 100% coverage of the pure core |
| bg3-corpus | 186k-pair EN–UK parallel corpus built from game localization files; MT evaluation across 5 metrics (BLEU, METEOR, TER, ChrF++, BERTScore). BA thesis |
| longcat_discord_bot | Production-style Discord bot: LLM chat with a tool loop and long-term memory, moderation, reminders. SQLite, 213 tests |
| tg-corpus | Telegram channel → clean, deduplicated text corpus. Two-stage design: API fetch, then a fully offline cleaning pipeline with auto-detected sign-off stripping |
| ua-freq-dict | Frequency dictionaries for Ukrainian: wordforms, lemmas, POS, n-grams (pymorphy3). Exports to CSV / XLSX / SQLite |
| tf-idf-summariser | Extractive TF-IDF summarization for Ukrainian and English — deterministic, no model downloads |
Every repository: bilingual README (EN / UK), MIT license, tests in CI.