Iuliia Zaitova

I am Iuliia (you-lee-yah) Zaitova, an AI Research Engineer at Diabolocom and a PhD candidate in Computational Linguistics at Saarland University (thesis submitted, awaiting defense). I have 5+ years of hands-on experience in applied machine learning for natural language and speech processing. I received an M.Sc. in Language Science and Technology at Saarland University (2019–2022; thesis: Modeling Mutual Spoken Word Intelligibility between Slavic languages) and a B.A. with Honours in Linguistics at Udmurt State University (2014–2018).

News

Work

AI Research Engineer, Diabolocom, June 2026–present. I benchmark open multilingual TTS systems (Qwen3-TTS, MOSS-TTS, Higgs Audio) for call-center synthesis, focusing on dialogue-turn consistency, voice cloning, and inference-time pronunciation correction for named entities. I am extending language coverage with supervised fine-tuning, followed by RL-based post-training.

Previously, Applied Scientist Intern, Amazon (Alexa TTS), July 2025–January 2026. I post-trained a production multilingual TTS model via SFT followed by RL, and designed the evaluation framework used to validate the resulting behavioral changes. I also built benchmarking datasets and automated diagnostics adopted by research and engineering teams for iterative model improvement.

Research

I post-train and evaluate transformer-based speech and language models using SFT, RL, and DPO, and I design the evaluation that checks whether those changes improve behavior, not just metrics. The work focuses on multilingual representations and on non-compositional and figurative language.

Selected work

idiom_s2t

Evaluation of speech-to-text systems (cascaded and end-to-end) on a hard case, idiomatic language. ACL 2025.

Paper Code

mwe-attention

Fine-tuned multilingual BERT models across tasks (POS, dependency relations, NER, topic classification) and languages, used to study how fine-tuning changes attention to multiword expressions.

Code Hugging Face

slavic_fixed_expressions

A cross-lingual inventory of non-compositional microsyntactic units in six Slavic languages.

Dataset

All papers are on the publications page.