邱钰迪

Key Skills

Software

Don't disclose

Top Subject Matter

LLM evaluation and red teaming

Manufacturing document parsing and engineering drawing AI workflow

AI video content generation workflow

Top Data Types

Text

Video

Image

Top Task Types

Red Teaming

Text Generation

Point/Key Point

Question Answering

Text Summarization

RLHF

Evaluation/Rating

Freelancer Overview

3 years of daily intensive cross-platform LLM evaluation and adversarial testing (GPT-4, Claude, DeepSeek, Gemini, and o. Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Don't disclose, Coze · n8n · Langflow · RPA · Feishu Bot · Multimodal Model APIs, and Veo · Seedance · Seedream · NANA · GPT-4o · Imagen 2 · ComfyUI · Midjourney. AI-training focus includes data types such as Text, Document, and Video and labeling workflows including Red Teaming, Function Calling, and Text Generation.

Labeling Experience

3 years of daily intensive cross-platform LLM evaluation and adversarial testing (GPT-4, Claude, DeepSeek, Gemini, and other models).

Don't discloseTextRed Teaming

Practiced structured adversarial evaluation of LLM behavior to detect subtle shifts, logical inconsistencies, hallucination patterns, and boundary failures across multiple model versions. Performed continuous model assessment to improve reliability and controllability of generated outputs using prompt and interaction strategies. Worked across leading open and closed models to validate performance and robustness under adversarial conditions. • Conducted RLHF-style preference ranking and hallucination audit activities as part of LLM evaluation. • Built and tested adversarial prompt crafting and cross-model prompt architectures. • Designed evaluation approaches to measure behavior shifts and version-to-version changes. • Assessed prompt quality and controllable generation behavior for workflow deployment.

2023 - Present

Cross-border e-commerce AI video content production pipeline build (scripting, storyboarding, generation, output).

VideoText Generation

Built an end-to-end AI video content production pipeline for the US TikTok market, including scripting, storyboarding, and generation of video outputs. Designed multi-model prompt architecture to maintain consistent characters and scenes across generation steps. Orchestrated multiple generative model tools to produce structured creative outputs for reliable downstream publishing. • Generated video scripts and storyboards using prompt-driven LLM-based planning. • Authored and maintained multi-model prompt architectures for consistency. • Coordinated outputs from multiple video/image generation models. • Designed an integrated pipeline from ideation to final video output.

2026 - 2026

Manufacturing AI workflow implementation for steel structure manufacturing (full AI workflow deployment).

DocumentFunction Calling

Implemented an end-to-end AI workflow that parsed engineering/manufacturing documents and analyzed engineering drawings to drive process automation. Validated extracted information by orchestrating multimodal model APIs and workflow automation steps to support downstream production decisions. Reduced document analysis turnaround time by applying structured AI workflow design and evaluation. • Performed document parsing and engineering drawing analysis using multimodal model APIs. • Automated processing steps with RPA and workflow orchestration tools. • Designed workflow logic to support efficient engineering review and production process improvement. • Quantified impact via reduced analysis time and annual cost reduction.

2024 - 2024

Education

Work History