Sumin S.

Key Skills

Software

Don't disclose

Top Subject Matter

Video generation (generative AI)

Multimodal AI research (VLMs/LLMs)

LLM finetuning, post-training

Top Data Types

Text

Image

Document

Top Task Types

Fine-tuning

Function Calling

Freelancer Overview

LLM Optimization Team, OptAI — Research Intern (Finetuning ~1B Korean language model). Brings 3+ years of professional experience across legal operations, contract review, compliance, and structured analysis. Core strengths include Internal, Proprietary Tooling, and Don't disclose. Education includes Bachelor of Science, Yonsei University (2026). AI-training focus includes data types such as Text and labeling workflows including Fine-tuning, Evaluation, and Rating.

Labeling Experience

LLM Optimization Team, OptAI — Research Intern (Finetuning ~1B Korean language model)

TextFine-tuning

Finetuned a ~1B-parameter Korean language model for the LG U+ call summarization project as part of the LLM Optimization Team. The work involved preparing training data and driving the model adaptation workflow toward better summarization performance. In addition, the experience supported downstream evaluation and deployment planning for an on-device demo. • Korean language model finetuning for call summarization • Training-data preparation and optimization workflow • Support for model evaluation and iteration • Preparation for on-device deployment demonstration

2025 - 2025

Know Yourself : Building an LLM with Strong Knowledge of AI — Project (QA benchmark + finetuning)

TextFine-tuning

Developed an AI-knowledge QA benchmark and fine-tuned a small language model on a domain-specific corpus for the Know Yourself project. This required preparing or selecting training text data from the AI knowledge domain to support model adaptation. The benchmark and fine-tuning together enabled structured QA evaluation and improved response relevance. • Built AI-knowledge QA benchmark • Domain-specific corpus preparation for training • Fine-tuned a small language model • QA-oriented evaluation/iteration workflow

2024 - 2025

AkaLlama : Educational Korean LLM — Project (Dataset building + ELO evaluation)

Text

Built and refined a large-scale Korean LLM dataset tailored to Yonsei University students for the AkaLlama educational LLM project. Designed and deployed a pairwise comparison evaluation system using ELO ranking to assess relative quality of QA pairs. The work supported dataset curation, generation/evaluation of question-answer content, and model-quality measurement. • Large-scale Korean QA dataset building/refinement • Pairwise ELO-based evaluation for QA quality • Dataset tailored to university student context • QA pair assessment and iteration loop

2024 - 2025

Multimodal AI Laboratory, Yonsei University — Undergraduate Researcher

Don't discloseText

Conducted multimodal AI research on vision-language models (VLMs) and large language models (LLMs) through the Multimodal AI Laboratory. The role included building and validating multimodal systems, as well as collaborating on projects requiring dataset design and comparative understanding. Work specifically covered multimodal UI/UX understanding efforts, which typically require curated examples, evaluation sets, and performance assessment. • Research on VLMs and LLMs • Dataset/model evaluation and collaborative experimentation • Visual text design transfer related work • Multimodal UI/UX understanding support

2023 - 2025

WorkMate : Developing an AI Legal Chatbot for Foreign Workers — Project (RAG system)

TextFunction Calling

Developed a database and retrieval-augmented generation (RAG) system for a legal chatbot intended to support foreign workers. The project involved assembling and organizing text knowledge for retrieval, which is central to RAG training/evaluation workflows. It also required creating prompt/response behavior for legal QA interactions using retrieved context. • Built legal chatbot knowledge base for retrieval • Implemented RAG system for domain QA • Organized text data for retrieval augmentation • Supported legal-domain QA behavior design

2024 - 2024

Education

Work History