Hongcheng H.

Key Skills

Software

Label Studio

Top Subject Matter

Economic Text Annotation

NLP Date Classification

Business Marketing Data Processing

Top Data Types

Text

Image

Video

Top Task Types

Classification

RLHF

Evaluation/Rating

Data Collection

Prompt + Response Writing (SFT)

Cuboid

Freelancer Overview

Project Leader. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Education includes Bachelor of Business Administration, 北师香港浸会大学 (2025) and High School Diploma, N/A (2025).As a first-year Applied Economics major and team leader for our “Sanchuang” smart sleep mask project (integrating soundproofing and light-blocking features), I led data collection and integration across multiple channels. We gathered user pain points from 200+ survey responses, online reviews, and competitor product specs. I manually labeled each data entry into categories such as “noise sensitivity,” “light leakage,” “comfort,” and “price expectation.” To ensure annotation quality, I designed a simple validation workflow: after two team members independently labeled a subset, I used a GPT‑based prompt to cross-check inconsistencies and flag ambiguous cases. This AI-assisted review reduced manual verification time by 40% and improved inter-annotator agreement from 78% to 91%. My role also involved integrating multimodal inputs — including user text feedback, decibel test logs, and material datasheets — into a structured dataset that guided our product design iterations. What makes my experience stand out is my focus on cost reduction and efficiency gain through AI tools. Instead of hiring extra annotators, I built a lightweight AI checking pipeline: I uploaded our labeled samples to a multimodal AI (GPT-4 with vision) to verify image-text alignment (e.g., matching user-uploaded sleep environment photos with their noise complaints) and to detect logic conflicts (e.g., a user requesting both “ultra-thin” and “maximum noise blocking”). This automated quality check cut manual review hours by over 50% for a team of four. As the leader, I coordinated weekly syncs, assigned data tasks based on each member’s strengths, and maintained a shared error log. My economics training helps me quantify trade‑offs — for example, calculating that a 10% improvement in label accuracy could save 15 hours of downstream model debugging. I am eager to apply this cross-disciplinary mindset and AI‑first efficiency approach to larger-scale data annotation and training projects. Feel free to adjust specific numbers or AI tools (e.g., replace GPT‑4 with Claude, etc.) to match your actual experience. Let me know if you need a shorter version or additional emphasis on any point.

Education

Work History