Context
BloGeek was a blog platform project built with React and Spring. My role was limited to the AI side, where the goal was to provide machine learning modules that could enrich text handling inside the broader product.
Problem
For a blog system to feel more expressive and adaptive, it needed more than standard CRUD or content management features. From the ML perspective, the team needed text models that could identify polarity and generate style variations that might support downstream content workflows and data augmentation.
Implementation
My contribution focused on two main NLP tasks. First, I worked on polarity recognition using a KoBERT-based classifier to determine whether text was positive, negative, or neutral. Second, I contributed to style transfer using a KoBART-based pipeline that generated stylistic variations of Korean sentences, which was also useful for data expansion. This work included dataset handling, training and inference code, and repeated experiments to improve the behavior of the models.
Outcome
The project provided working AI modules for polarity recognition and style transfer within a web application context. For me, it was valuable hands-on experience in contributing Korean NLP models to a team product built around React and Spring, while staying focused on the ML layer rather than the full-stack implementation.