All work

Turkish News NLP

Summarization and classification

Turkish News NLP

Short, medium or long. And what it is about.

News articleShortMediumLongCategory and confidenceEconomyTechnologySports
Illustration
Role
ML and full-stack engineering
Year
2025
Platform
Web app with REST API
Built with
  • PyTorch
  • Hugging Face Transformers
  • Flask
  • React
  • REST

The problem

Most summarization tools are built for English. Turkish is agglutinative, so a model has to be chosen and served with that in mind.

The idea

Serve two Transformer tasks behind one REST API: an abstractive summary at the length the reader picks, and a category with its confidence, so the interface can be honest about uncertainty.

How it’s built

  • Model serving

    PyTorch and Hugging Face models loaded once in a Flask service and exposed as REST endpoints.

  • Interface

    A React client to paste an article, switch summary length and see the predicted category with its score.

Where it is now

A full-stack app that summarizes and classifies Turkish news end to end.