XueyanZ

I am XYZ.
I worked in AI compiler for 3+ yrs.
I focus on smarter & faster models.

Application
└── Model
    └── Infra
        └── Hardware
            └── Resources

Publications (show selected / show all)

UORA: Uniform Orthogonal Reinitialization Adaptation in PEFT

Xueyan Zhang*, Jinman Zhao*, Zhifei Yang, Yibo Zhong, Shuhao Guan, Linbo Cao, Yining Wang

Annual Meeting of the Association for Computational Linguistics (ACL 2025)

anthology | arXiv

Tiny Budgets, Big Gains: Parameter Placement Strategy in PEFT

Xueyan Zhang*, Jinman Zhao*, et al.

Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)

anthology

Large Language Model is not a (Multilingual) Compositional Relation Reasoner

Xueyan Zhang*, Jinman Zhao*

First Conference on Language Modeling (COLM 2024)

arXiv | openreview

Projects

OpenTikZ

Open-source TikZ figure library & AI agent skill for academic papers

The "Flaticon for academic TikZ" — describe a figure (system diagrams, neural-net architectures, pipelines) and get paper-ready, editable TikZ back.

Tech Stack = TikZ/LaTeX + Claude Code Skill + GitHub

iMarkAi.com

Agentic bookmark with semantic search & web interactions

iMark bookmark = RAG + Agentic Tab Automation > Chrome Extension

Tech Stack = Vector DB (Pinecone) + LLM (Google Gemini) + Payment (Stripe) + Users (Supabase)

ArchiValley.AI

Web-based AI architectural renderer

Industry Impact: ArchiValley addresses the slow and expensive bottleneck of traditional architectural visualization. This AI-powered platform transforms 3D models and rough sketches into high-fidelity renders in seconds.

Tarot: Starry Guidance

Personalized Tarot Reader on WeChat

A full-stack mobile mini-program that combines traditional tarot with modern LLM interpretation.

Rapid Growth: Developed and launched in under a month, attracting over 500 users in the first 90 days.