Sujin Kim

Key Skills

Software

Appen

Other

Top Subject Matter

LLM evaluation in Korean & English

Data annotation in Korean & English

Video/text quality advice

Top Data Types

Image

Text

Video

Top Task Types

Bounding Box

Polygon

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I have extensive experience in data annotation and AI training data, including OCR annotation for both Korean and English in media contexts with RWS Group. Additionally, I contributed to an Interactive Recommendation project with Appen, focusing on enhancing top search results and refining video search functionality. My work also includes handling complex polygon annotation projects for medical purposes, as well as annotating Korean audio data. Furthermore, I have worked as a Korean linguist on a Sentence By Sentence Level Factuality project (Korean/English) with Outlier. These experiences have equipped me with a keen attention to detail, the ability to manage multilingual datasets, and a strong understanding of how to optimize data for AI training and machine learning models.

Labeling Experience

Outlier Sentence By Sentence Level Factuality Evaluation Task

OtherTextTranslation LocalizationEvaluation Rating

Scope: This task focuses on evaluating model-generated responses to prompts, assessing the factual accuracy of the claims made in each response. The goal is to ensure that the information provided by the model aligns with verified data and facts. Data Labeling Tasks: Tasks include identifying factual claims within the model's responses, assessing the accuracy of these claims, and conducting internet research to provide supporting or contradicting URLs for each claim. Project Size & Duration: The project is ongoing and requires flexible engagement based on the number of prompts and responses evaluated, typically involving several hours of work each week. Quality Measures: Quality is maintained through thorough evaluation of factual accuracy, with requirements for supporting evidence and appropriate citations to ensure reliability and credibility of the claims assessed.

Scope: This task focuses on evaluating model-generated responses to prompts, assessing the factual accuracy of the claims made in each response. The goal is to ensure that the information provided by the model aligns with verified data and facts. Data Labeling Tasks: Tasks include identifying factual claims within the model's responses, assessing the accuracy of these claims, and conducting internet research to provide supporting or contradicting URLs for each claim. Project Size & Duration: The project is ongoing and requires flexible engagement based on the number of prompts and responses evaluated, typically involving several hours of work each week. Quality Measures: Quality is maintained through thorough evaluation of factual accuracy, with requirements for supporting evidence and appropriate citations to ensure reliability and credibility of the claims assessed.

2024

Ads Quality Rating (global)

OtherTextEvaluation Rating

Scope: This project involved evaluating the quality and relevance of advertisements to ensure they meet specific guidelines and user expectations. The aim was to enhance ad effectiveness and improve user experience through accurate assessments. Data Labeling Tasks: Tasks included reviewing ad content, scoring relevance, and providing feedback on various advertising elements based on predefined criteria. Project Size & Duration: The project was conducted over a defined period, requiring flexible engagement based on task availability, typically involving several hours of work each week. Quality Measures: Maintained high standards of accuracy through consistent evaluation, with guidelines and performance metrics in place to ensure quality assessments.

Scope: This project involved evaluating the quality and relevance of advertisements to ensure they meet specific guidelines and user expectations. The aim was to enhance ad effectiveness and improve user experience through accurate assessments. Data Labeling Tasks: Tasks included reviewing ad content, scoring relevance, and providing feedback on various advertising elements based on predefined criteria. Project Size & Duration: The project was conducted over a defined period, requiring flexible engagement based on task availability, typically involving several hours of work each week. Quality Measures: Maintained high standards of accuracy through consistent evaluation, with guidelines and performance metrics in place to ensure quality assessments.

2024

Interative Recommendation

AppenVideoEntity Ner ClassificationRelationship

Title: Interactive Recommendation - Prompt & Recommended Content Relevancy / Prompt Understanding Labeling Scope: This project aimed to enhance the accuracy of search results and video recommendations by analyzing user intent in Korean. It focused on delivering relevant content that meets user needs, improving their overall search experience. Data Labeling Tasks: Tasks included annotating search prompts and evaluating the relevance of recommended content to ensure accurate alignment with user queries. Project Size & Duration: Spanning over 6 months, this large-scale project required continuous data annotation for Korean-language content. Quality Measures: Maintained a 93% accuracy target, with regular quality checks and adherence to set working hours for consistent progress.

Title: Interactive Recommendation - Prompt & Recommended Content Relevancy / Prompt Understanding Labeling Scope: This project aimed to enhance the accuracy of search results and video recommendations by analyzing user intent in Korean. It focused on delivering relevant content that meets user needs, improving their overall search experience. Data Labeling Tasks: Tasks included annotating search prompts and evaluating the relevance of recommended content to ensure accurate alignment with user queries. Project Size & Duration: Spanning over 6 months, this large-scale project required continuous data annotation for Korean-language content. Quality Measures: Maintained a 93% accuracy target, with regular quality checks and adherence to set working hours for consistent progress.

2024

RWS OCR Text Annotation

OtherImageBounding BoxPolygon

Title: RWS OCR Text Annotation - Korean Language - Political Ads Scope: This project focused on marking OCR-detected text within images and verifying the accuracy of the language annotations. It involved human review to ensure that the majority of text in each image matched the target language, Korean, after initial automated recognition. Data Labeling Tasks: Annotated OCR-extracted text and confirmed language accuracy, working through separate language queues for precise verification. Tools: Performed annotations using the client’s platform, HALO. Project Size & Duration: This project spanned 5-6 weeks, with a focus on reviewing and annotating OCR-extracted text in Korean. It required a commitment of 4-5 hours per day

Title: RWS OCR Text Annotation - Korean Language - Political Ads Scope: This project focused on marking OCR-detected text within images and verifying the accuracy of the language annotations. It involved human review to ensure that the majority of text in each image matched the target language, Korean, after initial automated recognition. Data Labeling Tasks: Annotated OCR-extracted text and confirmed language accuracy, working through separate language queues for precise verification. Tools: Performed annotations using the client’s platform, HALO. Project Size & Duration: This project spanned 5-6 weeks, with a focus on reviewing and annotating OCR-extracted text in Korean. It required a commitment of 4-5 hours per day

2024 - 2024

Education

Work History