Zhu J.

Key Skills

Software

Other

Top Subject Matter

Architecture/Building Design

AIGC / Generative AI

Design & Creative

Top Data Types

Image

Top Task Types

Classification

Text Generation

Question Answering

Text Summarization

Fine-tuning

Freelancer Overview

With a background that combines architectural design and AIGC technologies, I have strong experience in transforming complex business requirements into AI-executable workflows. I am familiar with AI training data organization, annotation system design, and prompt engineering methodologies. Through extensive use of ComfyUI, Stable Diffusion, and Midjourney, I have built and optimized generative AI workflows in real-world projects, gaining hands-on experience in data cleaning, style consistency control, dataset selection, and LoRA fine-tuning. I am able to optimize training data quality and improve model output stability and consistency based on specific business objectives. In multiple AI-assisted design projects, I was responsible for building end-to-end AI generation pipelines from scratch, including data classification, annotation standards, style sample selection, training evaluation, and iterative optimization. I possess strong cross-disciplinary understanding and can quickly translate business and creative requirements into standardized AI workflows. Compared with traditional data annotators, I bring a stronger perspective on optimizing training datasets and prompt systems from both the generation quality and real-world application standpoint.

Labeling Experience

Architectural Facade Style & Typology Image Dataset for LoRA Fine-Tuning

OtherImageFine-tuning

Curated and annotated a 12,000+ image dataset of architectural facades and building typologies to support Stable Diffusion LoRA fine-tuning for AI-generated architectural concept design. Established a multi-label classification taxonomy covering 6 style categories (modernist, parametric, neo-classical, vernacular, brutalism, sustainable/green) and 8 building typologies, with cross-tagging for materials, scale, and lighting conditions. Designed an iterative annotation pipeline using ComfyUI-assisted pre-labeling with human-in-the-loop verification. Implemented granular tagging protocols — from macro-level style classification down to micro-level architectural detail annotation (fenestration patterns, material textures, structural articulation). Achieved 96% inter-annotator agreement through rigorous quality control rounds, directly contributing to a 40% improvement in LoRA model output accuracy and stylistic coherence on downstream AI-generated design proposals for the Sichuan Seed Industry Innovation Complex project.

2025 - 2025

Education

Work History