István ÜvegesIstván Üveges
ML • NLP • XAI

Machine Learning for real-world impact

NLP, Plain Language & AI research with practical impact.

From research to applications, my focus is on clarity, reliability, and results that matter.

Embedding visualization of plain vs original texts

How plain and non-plain sentences separate in embedding space.

Portrait of István Üveges

About me

I am István Üveges, a computational linguist and Machine Learning researcher with a PhD in linguistics and a BSc in computer science. I specialize in Natural Language Processing (from classical models to Large Language Models) with a strong focus on clarity, interpretability, and practical impact. I work with organizations to turn text data into actionable insights, building solutions that are transparent, reliable, and tailored to real-world needs. Alongside research and applied projects, I write professional blog posts to make AI and NLP more accessible to a broader audience.

What Machine Learning Means in Practice

Machine learning is a practical tool for making sense of data. From solid baselines to cutting edge language models, I apply a wide range of methods with one goal in mind: clarity and understanding.

Classical ML

Solid baselines, clear starting points.

Classical ML

Classical methods are a reliable way to start. They set clear benchmarks and show when it is worth moving to more advanced models.

Transformers

Modern NLP at full power.

Transformers

Transformers capture patterns that older models miss. They make sense of complex texts and open the door to domain specific solutions.

Explainability

No black boxes, only clarity.

Explainability

AI should be transparent, not mysterious. With interpretability tools we can see how models decide and why their outputs matter.

Data Augmentation

When real data is not enough.

Data Augmentation

Real data can be scarce or unbalanced. Augmentation fills the gaps with synthetic examples that improve accuracy and fairness.

Unsupervised Learning

Let the data speak.

Unsupervised Learning

Unsupervised methods uncover hidden structure. Clusters and maps reveal themes that are invisible without labels.

Evaluation & Benchmarking

Fair tests, real insights.

Evaluation & Benchmarking

Not every model performs the way it claims. Careful evaluation shows what works, what fails, and where progress is real.

Hyperparameter Optimization

Tiny tweaks, big gains.

Hyperparameter Optimization

Performance often depends on fine details. Careful tuning can turn an average model into a strong and reliable solution.

Multilingual NLP

Language knows no borders.

Multilingual NLP

People communicate in many languages. Multilingual models break down barriers and connect ideas across cultures.

Model Robustness

Strong, fair, reliable.

Model Robustness

Good models perform well in easy cases. Robust models stay accurate under stress and remain fair in sensitive contexts.

Let’s Work Together

Practical NLP and Machine Learning solutions tailored to real-world text challenges. From rapid prototyping to audits and explainability, I focus on clarity, robustness, and impact.

ConsultationModel audit / XAIProof-of-ConceptData pipelinesTeaching / Talks

I usually reply within 24 hours. Currently at HUN-REN CSS • POLTEXTLAB