Decision Analytics + AI @ HKUST

Choi Cheuk Man

Building systems at the intersection of operations research, machine learning, and data analytics.

OptimizationMachine LearningData AnalyticsOperations ResearchStatistical ModellingSimulationPredictive AnalyticsPrescriptive AnalyticsSupply Chain AnalyticsData VisualizationDecision ModellingRisk Assessment
About

A bit about me

I'm a final-year student at the Hong Kong University of Science and Technology, majoring in Decision Analytics with an Extended Major in Artificial Intelligence. My skillsets are at the intersection of optimization, machine learning, and data analytics.

My Final Year Project reimagined hospital logistics with a hybrid LLM + OR-Tools pipeline, cutting porter pickup times by 35.8%. Beyond that, I've fine-tuned LLMs on HPC clusters, built AI-driven web applications, and analyzed physiological signals during a research exchange at DTU in Denmark.

I'm graduating in December 2026 and looking for roles where analytical precision drives real impact, whether in AI engineering, consulting or supply chain optimization.

Work

Selected projects

HKUST Final Year Project

AI-Driven Porter Dispatch System

A hybrid LLM + operations research system that optimizes real-time hospital porter logistics. DeepSeek LLM parses bilingual natural-language requests, Google OR-Tools solves a VRPTW assignment, and a RAG policy advisor flags KPI risk, all built on 81,349 real hospital records.

0.0%pickup-time reduction vs greedy baseline
PythonFlaskGoogle OR-ToolsDeepSeek LLMRAGpandasTailwind
AI-Driven Porter Dispatch System, figure 1
AI-Driven Porter Dispatch System, figure 2
LLM Fine-Tuning & Serving Systems preview
LLM Systems Engineering

LLM Fine-Tuning & Serving Systems

LoRA fine-tuned Qwen3-1.7B on HKUST's HPC4 cluster (A30 GPUs, SLURM). Ran data-scaling and rank ablations on Amazon Reviews. Benchmarked vLLM vs Ollama serving, achieving a 15.29× throughput speedup with batched inference. Evaluated Vidur scheduler comparison (vLLM vs Sarathi).

0.00×throughput speedup (batched vs sequential)
PyTorchHuggingFaceLoRA/PEFTvLLMSLURMHPC
Lunacourt preview
AI Conflict Mediation App

Lunacourt

A web app where an AI persona mediates disputes, helping both parties feel heard. Supports case submission, AI-generated verdicts, appeals, ratings, and case statistics. Built as a PWA with real-time session management.

PWAreal-time AI mediation app
Next.jsTypeScriptTailwindDeepSeek AITurso/SQLiteDrizzle ORMPWA
HRV Physiological Synchrony Analysis preview
Data Analytics & Signal Processing

HRV Physiological Synchrony Analysis

Python signal processing package analyzing heart rate variability and inter-subject correlation across 91 participants during collaborative tasks. Computes RMSSD, SDNN, PSD via Welch's method, and ISC to study cardiovascular coupling during teamwork.

0participants analyzed
PythonNumPySciPymatplotlibpandasFFT/PSD
Wheel Options Scanner preview
Personal Investing Tool

Wheel Options Scanner

CLI tool that automates screening for The Wheel options strategy. Computes analytical Black-Scholes Greeks, maintains a rolling IV Rank cache, filters by delta/return targets, and flags earnings risk. Built for learning quantitative finance.

0tickers scanned
PythonBlack-Scholesyfinancepy_vollibIV Rank
Toolkit

Tools I work with

AI / ML

PyTorchHuggingFaceLoRA/PEFTvLLMRAGscikit-learn

Optimization

Google OR-ToolsGurobiVRPTW

Development

PythonSQLC++VBALaTeX

Infrastructure

SLURM / HPCVercelGit

Tools

Microsoft ExcelPower BI

Data & Analysis

pandasNumPySciPymatplotlibseaborn