Jujin J.

Key Skills

Software

Label Studio

Top Subject Matter

Healthcare – Medical Records & Patient Data

Technology – AI/ML & LLM Development

Finance – Risk Analysis & Fraud Detection

Top Data Types

Text

Document

Audio

Top Task Types

Classification

Text Summarization

Fine-tuning

Segmentation

Freelancer Overview

During my graduate studies in Artificial Intelligence and Business Analytics at Lingnan University, I developed deep hands-on expertise in AI training data curation and annotation. My most significant contribution was independently designing and building a 564-sample JSONL fine-tuning dataset from scratch for my MindJournal AI project — carefully crafting multi-turn therapeutic dialogues across 3 distinct clinical roles (CBT therapist, emotional support counselor, and crisis intervention specialist), with strict attention to tone consistency, response length calibration, and bilingual (English/Chinese) alignment. This dataset was used to perform Supervised Fine-Tuning (SFT) on GLM-4-Plus, directly and measurably improving the model's role adherence, output quality, and cross-lingual generalization — giving me firsthand understanding of how annotation decisions upstream translate into model behavior downstream. Beyond text labeling, I have hands-on experience with multimodal data pipelines, including speech-to-text (ASR) annotation workflows integrating GLM-ASR, Whisper, and Google Speech fallback chains, as well as handwriting OCR data processing via GLM-OCR. Through my coursework in Machine Learning, Deep Learning, and Big Data Analytics, I have worked extensively with structured and unstructured datasets — applying preprocessing, cleaning, deduplication, and quality validation using Python and SQL. My background as a System Verification Engineer at Streamax Technology further sharpened my eye for edge cases, data anomalies, and standardized QA processes. I am detail-driven, highly comfortable with repetitive precision work at scale, and experienced in following rigorous annotation guidelines to ensure dataset integrity that models can rely on.

Labeling Experience

LLM Fine-Tuning Dataset Curation for Mental Health Dialogue AI

TextClassification

Independently curated a 564-sample JSONL fine-tuning dataset for a LLM-powered mental health dialogue system (MindJournal AI). Designed and annotated multi-turn therapeutic conversations across 3 clinical roles (CBT therapist, emotional support counselor, crisis intervention specialist), with strict quality control on tone consistency, response length, and bilingual (English/Chinese) alignment. Dataset was used to perform Supervised Fine-Tuning (SFT) on GLM-4-Plus, directly improving role consistency and multilingual output quality. Additionally handled multimodal data annotation including ASR transcription validation and handwriting OCR processing

2026 - 2026

Education

Work History