Bi Yuxi

Fashion design / Brand systems / AI-native product research

From fashion design to Fashion AI research and product practice.

Foundation

Fashion design education and training across SCAD, Paris, and Milan.

Practice

Five years of senior fashion design and brand-building experience.

Extension

At Tongji University, I translate fashion knowledge into Fashion AI research.

01SCAD / Paris / Milan
025 years in fashion design practice
03StePO-Rec co-first author · EI indexed
04IQON-3000 Recall@1 +28.3%
View Selected WorkView Resume

Proof

Representative Achievements

A focused view of education, showcase experience, publications, and research contribution.

EducationView resume

SCAD · ECSCP · Tongji

SCAD Scholarship · SCAD GPA 3.54/4 · Tongji GPA 4.91/5

ShowcaseBackground

NYFW Supima 2014

SCAD's only selected design for the fashion show.

PublicationView paper

StePO-Rec

Personalized Fashion Recommendation Thesis · Co-first author · ACM MM · CCF-A · EI indexed

SurveyView paper

RAG Survey

Visual architecture contribution to a highly cited retrieval-augmented generation survey.

Selected Cases

Core Fashion AI Projects

Research → System → Product → Playful Branch

01Research foundation

StePO-Rec

Personalized fashion recommendation method design research.

knowledge-guided recommendation

Co-first author · ACM MM · Recall@1 +28.3% · MAP +32.8%

02System prototyping

FashionAgent

Fashion multi-agent system prototype.

multi-agent decomposition

Role-based decomposition · Explainable recommendation · Foundation for OUTFITTED

03Product convergence

OUTFITTED

AI fashion product design and product-flow convergence.

usable agent product

Product lead · Multi-agent product chain from need interpretation to memory

04Playful branch

包治百病

Luxury gifting decision mini program / playful branch.

vibe-coded side application

Independent build · Style quiz, brand matching, and pixel-fashion interface

Research Visuals

Visualization for complex research structure

Taxonomy → Timeline → Framework → Evaluation → Collaboration

How dense research can be compressed into visual information that is immediately legible.

Content distillation: identify the few structural ideas that matter inside dense papers, surveys, and method stacks

Research logic: organize timelines, paradigm differences, method relations, and reasoning paths into one navigable system

Visual expression: make diagrams feel both analytically clear and editorially resolved

01

Fragments

• Modular RAG-3
z_Modular RAG-3
• Modular RAG-总结图
z_Modular RAG-总结图
• Rainbow Potion
Rainbow Potion 7-needle
• Flare
flare
02

Paradigms

• RAG Paradigms
Comparison between the three paradigms of RAG
• Hierarchical Multi-Agent
Hierarchical multi-agent execution
• Dual Fine-Tuning
Dual fine-tuning
• Retriever FT
retrieverFT
• Decision-Support Workflow
Decision-support workflow
• U-NIAH
U-NIAH framework
• Iter-RetGen
ITER-RETGEN
03

Taxonomy

• Human-AI Evolution
Human-AI collaboration evolution
• Urban Planning Taxonomy
Task taxonomy of LLMs in urban planning
• Modular Case
Case of current Modular RAG
• Modular RAG-1
z_Modular RAG-1
• Cost quadrant diagram
Cost quadrant diagram
• Linear RAG Flow
Linear RAG flow pattern
• RAG_CASE图-英文-0103修改
z_RAG_CASE图-英文-0103修改
04

Reasoning

• RAG + Reasoning Gains
Advantages of Combining RAG with Reasoning
• TOC
toc
• Practical Guide
Practical Guide on RAG & Reasoning
• Spatial Scales
Different spatial scales in urban planning tasks
• From LLM to RAG Then to RAG + Reasoning
From LLM to RAG and then to RAG + Reasoning
• Synergy Optimization
Implementation and optimization of the synergy
• Modular RAG-框架图
z_Modular RAG-框架图
• Databricks
z_databricks
05

Evaluation

• loop
loop
• Co-creative Frameworks
Representative co-creative frameworks
• four evaluation dimensions
The four evaluation dimensions
• Modular RAG-2
z_Modular RAG-2
• RAG vs Optimization
RAG vs other model optimization methods
• OpenAI
z_OPENAI
• Three RAG Paradigms
Comparison between three RAG paradigms
• Assistant Architectures
Typical conversational assistant architectures
06

Architectures

• purpose of the synergy
The purpose of the synergy
• Post-Retrieval Branching
Post-retrieval branching flow pattern
• generatorFT
generatorFT
• Synergy Patterns
Patterns of Synergy between RAG and Reasoning
• RAG-Reasoning Timeline
Timeline of studies on RAG-reasoning synergy
• RAG flow in REPLUG
The RAG flow in REPLUG
• graphRAG
z_graphRAG
• typical pipeline of multi-m…
The typical pipeline of multi-modal fusion
07

Bias / Data

• Modular RAG-4
z_Modular RAG-4
• Bias Reassignment
Generative de-biased matching framework
• RAG Milestone 0821
z_RAG Milestone 0821
• Data Submission Flow
Data collection and user submission process
• timeline of existing RAG re…
A timeline of existing RAG research
• Conditional Flow
conditional flow pattern
08

Branches

• RRR
RRR
• Pre-Retrieval Branching
Pre-retrieval branching flow pattern
• SelfRAG
selfRAG
• Practical Settings
comprehensive and practical settings
• Naive vs Adv.
Cases of Naive RAG and Advanced RAG

Trajectory

From fashion form to system form

Fashion Design → Brand / System Thinking → AI Research → AI-Native Product Prototyping

My training began in fashion design, but I have always been more interested in how structure, relationships, and systems generate style. Over time, that concern expanded into research methods, knowledge modeling, and AI product prototyping.

01

Foundation

Fashion design training built a sharp sense for silhouette, material, color, and audience perception.

02

Luxury Practice

Luxury and couture work strengthened brand thinking, product judgment, and the ability to structure taste into repeatable systems.

03

Research Shift

Graduate research translated those instincts into knowledge engineering, retrieval, reasoning, and explainable recommendation methods.

04

Current Synthesis

Current product work turns that combined background into AI-native tools and interfaces that feel both domain-aware and usable.