汪恒

Key Skills

Software

Label Studio

Top Subject Matter

Creative Writing/LLM Hallucination Auditing

Audio Transcription & ASR Quality Control

Cultural/Financial Text Entity Markup

Top Data Types

Text

Audio

Top Task Types

Transcription

Entity (NER) Classification

Freelancer Overview

Creative Writing Evaluator/LLM Hallucination Auditor. Core strengths include Label Studio and Audacity. Education includes Bachelor of Arts, 南京传媒学院. AI-training focus includes data types such as Text and Audio and labeling workflows including Evaluation, Rating, and Transcription.

Labeling Experience

Audio Transcription QA Specialist

AudioTranscription

I performed meticulous transcription and annotation for complex audio data from media interviews and entertainment podcasts to improve ASR model robustness. Responsibilities included quality assurance on over 100 hours of audio, correction of dialects, jargon, and overlapping speech, and reducing word error rates (WER) through manual intervention. I achieved over 98% transcription accuracy by aligning speakers and conducting segmentation annotation for multimodal training support. • Transcribed and QA'd 100+ hours of media audio content. • Corrected professional terminology, dialects, and overlaps in audio files. • Implemented precise speaker-alignment and segmentation tasks. • Enhanced ASR training corpora for multi-modal use cases.

2022 - 2023

Creative Writing Evaluator/LLM Hallucination Auditor

Label StudioText

In this role, I participated in large language model (LLM) creative content fine-tuning projects with a focus on artistic creation and cultural knowledge evaluation. I assessed over 3,000 AI-generated narrative texts for tonal consistency, logical coherence, and artistic aesthetics using a 5-level quantitative rating scale. Additionally, I designed and tested more than 500 adversarial prompts to probe model hallucination patterns related to traditional Chinese cultural concepts and provided optimization suggestions. • Evaluated thousands of AI-generated texts for quality and alignment. • Authored adversarial prompts to test model robustness in cultural scenarios. • Constructed SFT datasets with high cultural density for enhanced model output. • Provided RLHF, CoT, and prompt evaluation expertise for multi-perspective model auditing.

2022 - 2023

Text Data Annotator (NER & Classification)

Label StudioTextEntity (NER) Classification

I led structured data annotation of large-scale unstructured cultural and financial texts focusing on named entity recognition (NER). Over 10,000 data points were tagged, including cultural figures, art genres, and legal entities, maintaining over 95% inter-annotator agreement. My work supported model safety red-teaming via text classification of social media content for sentiment and risk factors. • Conducted NER tagging on tens of thousands of domain texts. • Maintained high annotation consistency in a collaborative setting. • Classified social media for sentiment and risk in red-teaming audits. • Enhanced structured training data for culture and finance models.

2021 - 2022

Education

Work History