PORTFOLIO LOG / 2026
An undergraduate background in Port, Waterway and Coastal Engineering, paired with freight-forwarding and marine cargo insurance experience—now developed through data analysis, simulation, and operations research.
01 / BACKGROUND
My engineering training began with ports, waterways, and coastal systems: the physical infrastructure that makes maritime trade possible. A later freight-forwarding internship exposed a different layer of the same system— cargo operations, insurance decisions, uncertainty, and commercial trade-offs.
That connection now shapes my interests in maritime digitalisation, industrial and systems engineering, supply-chain resilience, and intelligent transportation. I am building the quantitative foundation needed to move from observing operational problems to modelling and improving them.
02 / FEATURED PROJECT
A risk-aware decision-support prototype for maritime cargo insurance and intermodal route planning.
DECISION QUESTION
HarborShield combines a transparent risk model, Monte Carlo simulation, joint batch CVaR, whole-container allocation, and independent synthetic evaluation against four policy alternatives. It also displays public vessel-arrival data from the Maritime and Port Authority of Singapore.
RESEARCH INTEGRITY
Official and simulated inputs are separated. Vessel-arrival volume is not mislabelled as congestion. Model coefficients are not presented as actuarial rates until real claims data can support calibration.
View source registerv0.5 / REAL-DATA FORECAST TEST
Real AIS-derived Singapore container-ship port calls from IMF PortWatch. Train on 2022, select the method on 2023, then test on 366 days in 2024. The selected 28-day mean has MAE 3.29 calls/day versus 4.80 for yesterday's count; the fixed ridge model does not beat this simple baseline.
Read the real-data evidence 中文真实数据实验Source snapshots and reproducible tests are public. Port calls are activity indicators, not measured waiting times or cargo losses. Historical records were retrieved later; no live-deployment performance or financial benefit is claimed. AI-assisted implementation.
v0.4 / BILINGUAL PUBLIC-EVENT CASE
A Singapore 2024 public-event reconstruction with a constructed replenishment order. Adjust delay, inventory, payment terms and funding assumptions in your browser. Compare logistics cost, foregone contribution and funding interest, then inspect the cash ledger. Screen options against assumed funding, availability and contribution limits; no eligible option means no recommendation.
Explore the English business case 中文案例 English project brief ↗Public facts and constructed orders are separate. Document consistency and signature snapshots are included; the upload/signing lab runs locally. No company deployment, bank integration or realised savings is claimed. Built with AI-assisted coding.
v0.2 / EXPERIMENT EVIDENCE
Four disruption regimes, 10,000 held-out synthetic worlds per regime, five training seeds, paired uncertainty intervals, and shared-shock sensitivity. Baselines can coincide; the experiment does not assume that a more complex method wins.
Read the experiment evidenceThis is a static evidence page, not a hosted Python application. Built with AI-assisted coding; synthetic evaluation is not real-world validation.
03 / METHOD
Each layer answers a different question and remains visible enough to audit.
Estimate how cargo, packaging, transshipment, weather, and reliability affect claim probability.
Simulate thousands of cargo-loss and delay outcomes instead of relying on a single average.
Make cost, risk, time, and carbon priorities explicit through sensitivity analysis.
Assign whole containers under joint batch risk, then evaluate frozen decisions on separate synthetic samples.
04 / ACADEMIC DIRECTION
Port digitalisation, maritime data, intermodal logistics
Probability, systems modelling, optimisation, decision analysis
Supply-chain resilience, risk coordination, service trade-offs
Multimodal networks, disruption response, sustainable mobility
Trade-document integrity, digital signatures, issuer-key trust boundaries
Working capital, order contribution, liquidity timing and scenario sensitivity
05 / CONTACT
Code, methodology, data provenance, experiment results, and learning notes are published openly on GitHub.
Visit GitHub profile