林杨

Key Skills

Software

Doccano

Label Studio

Prodigy

Scale AI

Argilla

Top Subject Matter

NLP Text Annotation and Semantic Classification

LLM Response Evaluation and AI Training Data

Bilingual Content Review and Cross-cultural Communication

Top Data Types

Text

Image

Document

Top Task Types

Classification

Data Collection

RLHF

Segmentation

Polygon

Evaluation/Rating

Computer Programming/Coding

Freelancer Overview

I have practical experience in AI training data preparation, text annotation, and linguistic quality review through my CBS5502 NLP project. I worked with the Kaggle MBTI English corpus, where we processed 8,675 raw records into over 421,000 post segments and extracted target-word sentences for “think” and “feel.” For the “feel” subset, we identified 22,587 candidate sentences, retained 18,636 usable items, and built a manually reviewed gold set for semantic classification. My strongest fit is text-based AI training work, including classification, evaluation, data cleaning, and guideline-based review. I reviewed “feel” sentences by distinguishing experiential/state meanings from propositional/judgment meanings, confirming labels, correcting errors, excluding noisy or incomplete examples, and documenting ambiguous cases. With my background in computational linguistics, generative AI, bilingual communication, and design QA, I am detail-oriented, consistent with annotation guidelines, and able to handle nuanced language data for LLM evaluation and NLP training tasks.

Labeling Experience

Disambiguating “Feel” in MBTI Thinking vs. Feeling Types

TextEvaluation/Rating

In my CBS5502 final project, I worked on an NLP data preparation and manual annotation workflow using the Kaggle MBTI English corpus. The project processed 8,675 raw rows into over 421,000 post segments, extracted target-word sentences for “think” and “feel,” and produced cleaned datasets with traceable audit logs. For the “feel” subset, we identified 22,587 candidate sentences, kept 18,636 usable items, and created a balanced 400-row gold set for manual review. My annotation work focused on distinguishing A: experiential/state uses from B: propositional/judgment uses of “feel.” I followed detailed review guidelines, checked whether each sentence should remain in the gold set, confirmed or corrected labels, marked exclusions such as truncated fragments or non-target usage, and supported final quality control. This experience strengthened my ability to follow annotation standards, resolve linguistic ambiguity, document decisions, and produce reliable training data for AI/NLP systems.

2026 - 2026

Education

Work History