Leon Cheung

Leon Cheung (张泽鼎)

M.S. Student in Artificial Intelligence at Yunnan University

Focusing on AI, Software Engineering & Industrial Systems.

🔍 Actively seeking internship opportunities

About

I am an M.S. student in Artificial Intelligence at Yunnan University (2025–2028) and a B.E. graduate in Vehicle Engineering from Guangdong University of Technology.

Before returning to graduate school, I worked for two years as a C++ software engineer, independently delivering industrial CAM and semiconductor equipment control systems. That experience gave me a strong foundation in systems programming, debugging, and shipping production software. Since then, I've expanded into AI engineering — building RAG systems, coding agents, and LLM-powered SaaS platforms from scratch.

I'm actively seeking internship opportunities in AI engineering or systems development, where I can bring both my industrial software experience and my recent AI project work to real-world problems.

Education

2025.09 – 2028.06 (Expected)

Yunnan University

M.S. in Artificial Intelligence

School of Information Science and Engineering. Focusing on computer vision and software engineering.

2018.09 – 2022.06

Guangdong University of Technology

B.E. in Vehicle Engineering

Experience

2025.04 – 2025.08

Han's Laser Smart Equipment Group Co., Ltd.

C++ Software Engineer

  • Developed Windows-based CAM software using Qt5; independently maintained the 3D general-purpose CAM project for aluminum window/door machining centers, delivering 10 releases.
  • Collaborated with QA, after-sales, and product teams to refine machining trajectory modules and added 20+ features including interval compensation.
  • Independently built an angle-steel CAM project from scratch based on product research, requirements, and design documents, delivering multi-user system, auto-nesting, and part file system integration.
2023.07 – 2024.09

Shenzhen Xinyichang Technology Co., Ltd.

C++ Software Engineer

  • Maintained and developed Windows-based industrial control software for semiconductor die bonders using MFC, covering motion control, vision inspection, dispensing control, and full HMI.
  • Integrated SECS/GEM protocol for MES communication and RS-232 serial communication with die bonder hardware.
  • Leveraged OpenCV, Halcon, and Matrox MIL vision libraries for pick-and-place and die-bonding vision capabilities.
  • Independently owned logging, i18n, database, multi-axis motion, multi-user, and UI subsystems.

Selected Projects

AGENT

CodePipe: A Locally-First Deterministic Pipeline Coding Agent

Coding Agent / BM25 / AST / Typer CLI / Docker / CI

  • Designed a deterministic coding agent pipeline that decomposes tasks into Gate, Locator, Generator, and Verifier stages, with LLMs handling only classification and patch generation to reduce hallucination and infinite-loop risks.
  • Implemented a closed-loop of localization, patching, and verification using BM25 + AST context trimming, SEARCH/REPLACE with difflib for patch application, L1/L2 validation, Git rollback, and Reflexion — 200 tests passed.
  • Shipped production-grade tooling including Typer CLI, REPL, multi-driver configuration, Docker sandboxing, and CI.
RAG

RepoRAG: RAG Assistant for GitHub Open-Source Repository Knowledge Bases

FastAPI / LangGraph / PostgreSQL + pgvector / Streamlit

  • Built a code-repository-aware RAG system with GitHub repo indexing, code-aware chunking, vector ingestion, Q&A API, and interactive frontend.
  • Implemented hybrid retrieval with trustworthy citations — Markdown heading-based splitting, Python AST function/class chunking, issue/PR semantic indexing, combined vector + full-text search with RRF fusion, and answers with GitHub permalinks, file paths, and line numbers.
  • Delivered production readiness with OpenAI-compatible/DeepSeek providers, Docker Compose, Alembic migrations, citation validation, refusal-to-answer for low-confidence queries, and offline evaluation covering Recall@k, MRR, citation coverage, and latency — 41 unit tests passing with ruff.
GEO

GEO Lens: Generative Engine Optimization Analytics Platform

Next.js / React / TypeScript / Tailwind CSS / Prisma

  • Delivered an AI search visibility SaaS workflow with project CRUD, five-dimensional GEO scoring, Q&A simulation, citation diagnostics, optimization suggestions, and report export.
  • Built a robust LLM integration layer with OpenAI-compatible client, DeepSeek/Mock backends, JSON mode, Zod validation, timeout trimming, session isolation, IP rate limiting, SSRF protection, and data sanitization.
  • Shipped the full-stack delivery loop — PostgreSQL data model, REST API, strategy library, readiness audit, and smoke tests covering the core path from project creation to report generation.
WX

Huazhou Dictionary — WeChat Mini Program for Dialect Pronunciation Lookup

WeChat Mini Program / JavaScript / glass-easel

  • Built a WeChat Mini Program for looking up character pronunciations in the Huazhou dialect (a Yue/Cantonese branch), supporting both Simplified and Traditional Chinese input with automatic script conversion.
  • Provides three pronunciation systems — Guangzhou Cantonese (Jyutping), Upper River, and Lower River Huazhou variants — with color-coded display for clarity.
  • Fully offline with zero cloud dependencies; dictionary covers 4,310 character entries with ~4,000 simplified-to-traditional mappings.

Notes