胡家瑞

Key Skills

Software

Don't disclose

Other

Top Subject Matter

Mandarin business content

product QA

LLM workflow testing

Top Data Types

Text

Video

Top Task Types

Prompt + Response Writing (SFT)

Freelancer Overview

Production Intern. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Don't disclose and Other. Education includes Bachelor of Engineering, Xi'an University of Technology (2026). AI-training focus includes data types such as Text and labeling workflows including Evaluation, Rating, and Prompt + Response Writing (SFT).

Labeling Experience

AI-Assisted Local-Merchant Content Generation Workflow | Product Testing / Entrepreneurship Practice | Remote | 2026 - Present

OtherTextPrompt + Response Writing (SFT)

Supported AI-assisted generation of promotional copy, scripts, and video-generation task prompts for local merchants. Validated that outputs matched merchant profiles, industry templates, video duration, tone requirements, available materials, and natural Chinese expression. Used structured prompt design and smoke-test thinking to identify missing constraints, field-mapping inconsistencies, task-state errors, and user-experience breakpoints. • Designed and tested structured prompts for content generation tasks • Performed constraint verification against merchant-specific requirements • Diagnosed workflow errors such as field mapping and task-state failures • Improved user-facing outcomes by refining prompt and QA criteria.

2026 - Present

AI Agent Workflow & LLM Output Evaluation | Independent Product QA Practice | Remote | 2026 - Present

Don't discloseText

Conducted Mandarin LLM response evaluation and product QA using Codex-style agents and ChatGPT to test instructions and outputs. Reviewed AI-generated Chinese business copy and scripts for instruction following, clarity, factual consistency, tone fit, template fit, and hallucination risk. Documented reproducible issues with symptom, expected vs actual results, reproduction steps, likely cause, and proposed fixes to produce actionable QA feedback. • Evaluated template selection and material/slot requirement satisfaction • Checked API-result consistency, task progress/result display, and user-facing error messaging • Applied response comparison and rule-based scoring with an RLHF-style preference judgment mindset • Followed Chinese annotation/labeling standards and maintained annotation consistency and error taxonomy.

2026 - Present

Education

Work History