Chen L.

Key Skills

Software

No software listed

Top Subject Matter

Multimodal LLM post-training

Alignment Domain Expertise

and dataset generation

Top Data Types

Video

Text

Image

Top Task Types

Fine-tuning

Data Collection

Entity (NER) Classification

Text Generation

Question Answering

RLHF

Text Summarization

Freelancer Overview

Multimodal LLM Algorithm Engineer, Huawei. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal and Proprietary Tooling. Education includes Master of Science, The University of Hong Kong (2028) and Bachelor of Engineering, South China University of Technology (2026). AI-training focus includes data types such as Image and Video and labeling workflows including Fine-tuning and Data Collection.

Labeling Experience

AI Data Annotation & Quality Specialist, ByteDance

VideoData Collection

Built an annotation system for video help-seeking data, including user intent, emotion detection, and scenario classification. Implemented a three-tier quality control workflow with auto-screening, manual review, and expert sampling to reach 98.5% accuracy. Produced large-scale edu-problem annotations with a knowledge tag taxonomy and difficulty grading for 50K+ problems. • Video annotation for intent, emotion, and scenario classification • Three-tier QC pipeline (auto-screening → manual review → expert sampling) • Knowledge tagging across Math/Physics/Chem and difficulty grading • Authored rules/training manuals and onboarded 15 annotators to improve efficiency by 25%

2026 - 2026

Multimodal LLM Algorithm Engineer, Huawei

ImageFine-tuning

Led multimodal LLM post-training data construction for image-text and video-text alignment, scaling to 100K+ samples per day. Built supervised fine-tuning datasets covering visual QA, image captioning, and video understanding tasks, enabling training of Qwen2.5-VL. Created RLHF preference datasets and performed DPO/PPO alignment to improve human preference win-rate by 12%. • Image-text/video-text alignment dataset construction • SFT dataset design for 20+ multimodal tasks • RLHF/DPO/PPO preference optimization using human preferences • Automated evaluation with bad case analysis to accelerate iteration cycles by 30%

2025 - 2026

Education

Work History