MXMinglang Xie

Researcher · Engineer · Builder

Minglang Xie

AI / Data Systems / Full-Stack Engineer

UNSW MPhil graduate working across graph machine learning, data systems, and production-oriented AI software. I care about systems that are technically rigorous, useful in practice, and honest about what they can do.

View projects Project progress Read articles GitHub Email

EDBT 2025 publication

Energy journal publication

Huawei big-data engineering

UNSW database tutor

Building AI SaaS products

Active projects

The products and tools currently moving.

A focused snapshot of current work across AI transcription, domain RAG, retail operations, recipe formulation, and map generation.

AI SaaSMusic AIFull-stack

MVP / quality iteration

Song2Notation

AI-powered music transcription for turning audio into staff notation, Jianpu, piano roll, MIDI, and portable notation formats.

Progress68%
Next.jsSupabasePython workerMusicXML
AI SaaSRAGFull-stack

Knowledge + retrieval iteration

Agarwood AI

A bilingual RAG knowledge assistant for agarwood education, fragrance guidance, cultural knowledge, and buyer consultation.

Progress55%
Next.jsSupabasePostgresOpenAI
SaaSRetail OpsFull-stack

MVP workflow closure

Space Display SaaS

A mobile-first SaaS/PWA for managing retail products, spaces, zones, shelves, layers, and placement search.

Progress62%
Next.jsSupabaseReact-KonvaZustand
Food TechToolsFrontend

Core tool polish

Gelato Recipe Generator

A formulation tool for building gelato and sorbet recipes with ingredient profiles, texture targets, metrics, warnings, and exports.

Progress78%
ViteReactTypeScriptLocal storage

Research

Selected publications

Research in graph community search and physics-informed photovoltaic forecasting.

01

EDBT 2025 · CCF-B

Deep Skyline Community Search

Minglang Xie, Jianye Yang, Wenjie Zhang, Shiyu Yang, Xuemin Lin

A learning-driven framework for efficient skyline community search in high-dimensional attributed networks.

  • Designed Skyline Learnable GCN (SLGCN) with dominance and link losses for unsupervised representation learning.
  • Proposed Skyline Community Score and Expected Score Gain for efficient online expansion.
  • Improved query efficiency by up to roughly three orders of magnitude at five attributes while maintaining comparable accuracy.
Read publication

02

Energy · SCI TOP journal

Assessing Solar-to-PV Power Conversion Models: Physical, ML, and Hybrid Approaches Across Diverse Scales

Caixia Li, Yuanyuan Xu, Minglang Xie, Pengfei Zhang, Bohan Zhang, Bo Xiao, Sujun Zhang, Wenjie Zhang, Xiaojing Hao

A cross-scale evaluation of physical, machine-learning, and physics-informed hybrid models for photovoltaic power conversion.

  • Contributed to a hybrid framework combining a time-aware multi-layer mixer with physical IV modelling.
  • Worked on implementation, tuning, data preprocessing, error analysis, and method and experiment writing.
  • The hybrid approach reduced RMSE relative to pure ML models in complex forecasting settings.
Read publication

Writing

Notes from the work

Long-form notes on research, systems, engineering, algorithms, and building AI products.

Engineering18 June 20261 min read

Building Song2Notation: From Audio to Sheet Music

A working note on the product and engineering decisions behind an AI music-transcription workflow.

AI SaaSMusic AI
Research10 June 20261 min read

What I Learned from Deep Skyline Community Search

Lessons from combining graph representation learning with a difficult high-dimensional search problem.

ResearchGraph ML