Leung Kwan Cheuk 
Quant Research Terminal
- 7 MCP servers over one curated data layer
- 36-arm walk-forward grid, 2005–2013 benchmark
- No-lookahead enforced by the pipeline itself
Leung Kwan Cheuk (Anson) — Info Security · AI · Hong Kong
Own the
whole stack.
I build web apps and AI pipelines — then I test them like an attacker.
Year-3 BSc (Hons) Information Security at PolyU. Full-stack web, agentic tooling, and LLM evaluation research — every project on this page ships with its evidence.
2×
LSCM AI INTERNSHIPS
6
SHIPPED PROJECTS
3
LANGUAGES · EN 中 普
01
Featured work
VideoMarketing — Automated Episode Pipeline
- One JSON spec in → finished video out
- TTS + headless Chromium + burned-in captions
- Loudness & delivery gates refuse bad builds
mathdse — HKDSE Maths Mock Generator
- Real HKDSE papers → structured LaTeX item bank
- Topic- and difficulty-profiled mock generation
- Distractor mapping turns wrong answers into diagnosis
This Portfolio — Next.js + CMS + 3D Hero
- Next.js 16 + Payload 3 — every word admin-editable
- Rigged 3D robot with cursor gaze, tiered by device
- Static tier ships zero 3D bytes to phones
acadwrite — Academic-Writing Analysis CLI
- EN + 中文 readability & lexical-diversity metrics
- 300+ automated quality-gate tests
- Reusable prompt library, CLI-first design
Self-Hosted Multi-Agent AI Environment
- 10 isolated agent profiles on Arch Linux
- Cost-based routing across multiple LLM providers
- Parallel research subagents with cited output
Other projects
Archive of everything else I’ve built — new entries land here first.
Archive02
Experience
Logistics and Supply Chain MultiTech R&D Centre · May 2025–Aug 2025
Project Research Assistant (Intern) — Customs Declaration Classification
- Improved an LLM customs-declaration classifier via pattern recognition & clustering
- Introduced t-SNE/UMAP visualisations for diagnosing classifier errors
- Flagged conflicting reference data before classification — better inputs, better outputs
Logistics and Supply Chain MultiTech R&D Centre · Jun 2026–Aug 2026
Project Research Assistant (Intern) — AI Education Platform & HS Classification
- Analysed LLM misclassifications across 5 HS chapters under a 6-category error taxonomy
- Measured HS6/HS10 accuracy per chapter with Python reporting scripts
- Reviewed ~50 AI-generated Chinese Q&A items; reported recurring fault patterns
- Benchmarked 9 HK education-AI services for client-facing reports
- Shipped React dashboard + bilingual features via Git/GitHub PR review
03
About
I’d rather show than tell. Every project here comes with a number — a passing test, a frame count, a log file — because “it works” doesn’t mean much until something is able to fail.
I’m in Year 3 of Information Security at PolyU. Outside class I build tools with an AI agent I set up myself, tutor maths, and produce short videos. I work in English, Cantonese or Mandarin, and I live in a Linux terminal.