The product needed ML components that could recognize user emotion and support a diary-generation workflow from daily conversation logs.
Project Detail
FRIMO: Conversational AI for Emotional Support and Diary Generation
A Korean conversational AI project connecting emotion recognition, chatbot, and summarization models to a diary-generation workflow.
Applied NLP and LLM Research Engineer Built project Summary available Conversational AIEmotion RecognitionKoBERTKoGPT2KoBARTNLP
Connects Korean NLP models into a user-facing AI pipeline for conversational product experience.
text
dataexperimenttraininginferenceevaluation
Centered the ML workflow around KoBERT-based emotion recognition and connected it to chatbot and summarization flows.
The project delivered an MVP-level conversational diary experience with an emotion recognition pipeline.
Context
FRIMO, short for Friend for Modern People, was designed as an application where users could talk with an AI companion and later receive an automatically generated diary summary of the day. The broader product combined chat, emotion-aware interpretation, and diary-style summarization for lightweight emotional support.
Problem
From the ML perspective, the system needed models that could interpret user emotion from conversational text and support downstream diary generation. This meant building practical model components that could operate within a larger application flow rather than as isolated research demos.
Implementation
My contribution focused on the machine learning part of the project, especially emotion recognition. I worked with a KoBERT-based classification pipeline, including training and inference code, corpus-based data handling, and repeated experiments for model optimization. I also contributed to the broader AI workflow by helping align the emotion recognition component with Korean chatbot and summarization models such as KoGPT2 and KoBART.
Outcome
The project reached an MVP stage with an ML pipeline that could support emotion-aware interaction and diary generation. For me, the most important result was gaining hands-on experience in building and integrating Korean NLP models in a user-facing conversational product.