Hz L.

Key Skills

Software

Other

Top Subject Matter

Enterprise document OCR and text detection

Shipping domain LLM instruction dataset

Enterprise and policy data semantic structuring

Top Data Types

Document

Text

Video

Image

Top Task Types

Object Detection

Prompt + Response Writing (SFT)

Action Recognition

Freelancer Overview

Enterprise Document Labeling and OCR Model Training. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal, Proprietary Tooling, and Other. Education includes Master of Science, 上海海事大学 (2026) and Bachelor of Science, 上海海事大学 (2023). AI-training focus includes data types such as Document, Text, and Video and labeling workflows including Object Detection, Prompt + Response Writing (SFT), and Action Recognition.

Labeling Experience

Instruction Data Labeling for Shipping Domain LLM Training

TextPrompt + Response Writing (SFT)

Participated in the construction of multidimensional instruction datasets by writing and curating prompt-response pairs across general knowledge, shipping domain, self-awareness, and safety ethics. Assisted in SFT pipeline development, injecting domain-specific and mixed-domain labeled data to guide LLM training. Collaborated with domain experts to review, annotate, and rate data quality for iterative improvements. • Designed instruction data labeling protocols for large language model fine-tuning. • Annotated shipping, ethical, and general instruction data for LLM training. • Performed quality assurance through expert review and iterative correction. • Ensured diversity and robustness in labeled instruction sets.

2026 - 2026

Enterprise Document Labeling and OCR Model Training

DocumentObject Detection

Responsible for constructing and labeling enterprise documents, such as contracts and invoices, to build a hybrid dataset for intelligent OCR model training. Optimized labeling strategies to improve the detection of complex perspective deformation and small targets using tailored sampling and label weighting techniques. Conducted iterative model and labeling refinements based on badcase analysis and performance feedback. • Built and labeled a dataset of enterprise documents for automated OCR. • Focused labeling improvements on challenging scenarios like perspective deformation. • Introduced advanced loss functions to increase text location accuracy. • Worked with a research team to improve object detection in documents.

2026 - 2026

Video Event Segmentation and Annotation for Multimodal LLM Research

OtherVideoAction Recognition

Labeled and curated video event segments using multimodal token fusion mechanisms for training models on event localization. Optimized segment labeling pipelines and generated event boundaries, ensuring temporal and descriptive accuracy for each video sample. Benchmarked and refined event labeling processes through comparative SOTA analysis and single-sample inference testing. • Conducted video event segment labeling and temporal boundary annotation. • Implemented multimodal prompt and token labeling strategies. • Enhanced dataset value for training and SOTA benchmarking. • Maintained annotation quality through systematic evaluation.

2025 - 2025

Prompt Engineering and Data Annotation for Policy Matching System

OtherTextPrompt + Response Writing (SFT)

Developed and labeled structured prompt templates to extract key attributes and semantic features from unstructured enterprise and policy texts. Applied DeepSeek LLM to align unstructured information into structured labeled data for policy matching system design. Evaluated and improved data labeling accuracy through validation experiments with real-world policy and enterprise datasets. • Built large-scale prompt templates for structured text extraction. • Labeled and aligned policy and enterprise data for semantic understanding. • Utilized DeepSeek and prompt engineering for high-fidelity data annotation. • Conducted performance validation on labeled dataset accuracy.

2025 - 2025

Education

Work History