Shaofeng G.

Key Skills

Software

Don't disclose

Top Subject Matter

Automotive speech transcription

Mandarin and Henan dialect speech corpus construction

LLM safety red-teaming and vulnerability evaluation

Top Data Types

Audio

Text

Video

Document

Top Task Types

Transcription

Red Teaming

Segmentation

Freelancer Overview

Data Annotation Specialist (audio transcription and segmentation) | Chinasoft International. Brings 10+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include ELAN, Don't disclose, and Praat. Education includes Associate Degree, Henan Vocational College of Applied Technology (2016) and Specialist Certificate, Beijing Good Project Smart Service Alliance (2021). AI-training focus includes data types such as Audio and Text and labeling workflows including Transcription, Red Teaming, and Segmentation.

Labeling Experience

Corpus Construction & AI Safety Specialist | Remote (AI red-teaming and safety data work)

Don't discloseTextRed Teaming

Performed adversarial testing and safety evaluation by generating and operationalizing attack prompts against LLMs. Documented and categorized vulnerabilities related to regional bias, NSFW content exposure, and privacy leakage to inform mitigation. Hardened model safety boundaries by developing optimized refusal-response behaviors for regulated generative AI scenarios. • Built and tested 200+ high-level attack prompts using logic sandboxing and dialect-nested slang • Conducted jailbreak detection and refusal-response hardening for compliance • Identified vulnerability categories including bias, NSFW, and privacy leaks • Ensured red-teaming activities were aligned with generative AI regulations

2024

Data Annotation Specialist (audio transcription and segmentation) | Chinasoft International

AudioTranscription

Led high-fidelity transcription and boundary alignment for automotive audio and speech datasets with tight timing control. Labeled Mandarin and Henan dialect segments by aligning acoustic cues (e.g., pitch/rhythm) to semantic intent context. Ensured annotation quality with an emphasis on precision and consistency across large daily volumes. • Transcribed and segmented audio with ±100ms boundary control • Annotated for complex intent patterns in Henan dialect speech • Applied structured multi-tier annotation using ELAN for interaction nuances • Maintained high accuracy (97.5%) for large-scale text/audio-related entries

2023

Corpus Construction & AI Safety Specialist | Remote (speech corpus annotation)

AudioSegmentation

Created end-to-end multimodal speech corpus annotations for next-gen speech models. Labeled complex irony and intent in Henan dialect by aligning acoustic markers such as pitch and rhythm with semantic context. Built structural annotation frameworks to capture interaction dynamics including interruptions and emotional shifts. • Led multi-tier corpus annotation to represent human-machine interaction nuances • Applied acoustic-semantic alignment to support intent alignment • Captured interruptions and emotional shifts as part of structured labels • Used ELAN to produce consistent annotation structure for training data

2022

Education

Work History