Research

A software engineer by profession, reading and building my way into research. The direction is NLP for low-resource languages - the Turkic family, such as Azerbaijani, and the West Slavic group, such as Slovak and Czech - and this page is where that work is kept while I prepare for a PhD in it.

Directions

1
NLP for Azerbaijani
An agglutinative language with little annotated data. What I want to understand: how far tokenisation and morphology-aware modelling can carry a language before corpus size becomes the wall.
Morphological segmentationCorpus buildingTransfer from high-resource languages
2
Transfer across the Turkic family
Turkish has data; Azerbaijani, Turkmen and the rest have far less. The family shares structure, so the question is how much of a high-resource sibling transfers, and where it quietly stops.
Multilingual pretrainingShared subword vocabulariesZero-shot evaluation
3
Czech and Slovak
Two close, morphologically rich languages I live between. They make a natural testbed for cross-lingual work: near-identical structure, separate data, separate communities.
Cross-lingual alignmentRich morphologyLow-resource fine-tuning
4
Adaptive learning systems
Where I started, and still one thread: modelling what a learner knows, deciding what to teach next, and explaining it in language grounded in the course material.
Bayesian Knowledge TracingPPORetrieval-augmented generation
in EduVision, LangVis