Taisiya Glushkova

Hi! I am a ML/NLP engineer working on LLMs, evaluation, and AI-powered products. I hold a PhD in Computer Science from Instituto Superior Técnico (SARDINE lab, 2024), where I was supervised by André F. T. Martins and Chrysoula Zerva. Currently building Chess Baby Steps, an AI-powered chess trainer. Based in Lisbon, Portugal.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github  /  LinkedIn

My doctoral research focused on the robustness of NLG evaluation models and on methods that quantify and communicate the uncertainty of their outputs. In particular, I worked on uncertainty-aware MT evaluation, aiming for trustworthy systems that are robust to the variation observed across different languages and domains. I co-authored widely used COMET variants (COMET-22, CometKiwi), with 1000+ citations on the corresponding EMNLP/WMT publications.

After the PhD I moved to the applied side. I worked on adapting LLMs for low-resource language pairs, and currently I'm building my own AI products end-to-end (such as Chess Baby Steps) and evaluating LLM-based systems in production.

Before joining IST, I received a MSc and BSc in Computer Science at CS HSE in Moscow, Russia.

When not at my computer, I enjoy playing chess!

An AI-powered chess trainer that turns your losses into personalized practice: it pinpoints your losing patterns and generates custom puzzles, opening curricula, and tactical drills from positions you misplayed in your own games. Built end-to-end solo (LLMs + Stockfish, full-stack web app), and developed in the open with a strong feedback loop from my TikTok community.