Projects
My final year project on trust-aware LLM agents, and the undergraduate research I have been doing at HKUST on how the built environment shapes the heat a person is actually exposed to.
TrustShop — Final Year Project
Jul 2026 – Present · supervised by Prof. Nevin Lianwen ZHANG
What it is
TrustShop is an agent-orchestrated online shopping assistant. The premise is that product reviews are useful in bulk but unreliable individually, so the system treats trust as a first-class signal rather than assuming every review deserves equal weight. An agent layer decomposes an open-ended shopping question into steps, decides which tools to call, and routes evidence according to how trustworthy its source appears.
My contribution
- Trust-weighted review summarisation — designed and implemented a module using aspect clustering, pros/cons extraction, and self-critique verification with LLMs.
- Data acquisition — built a deterministic pipeline with schema validation and commercial API fallback, collecting and processing product and review data across platforms.
- Agent orchestration — developed tool schemas and contributed to the orchestration layer for open-ended query decomposition and trust-aware routing.
Areas
Heat Index Estimation — UROP
Jun 2025 – Present · Undergraduate Research Opportunities Programme, HKUST
Supervised by Prof. FUNG, Jimmy Chi Hung; co-supervised by Dr. CHAN, Jimmy Wai
Man, and Dr. Tilman Leo Hohenberger (from 2026 Summer).
The problem
A weather station measures temperature at one single point. Two spots a street apart can still feel very different, because one sits in direct sun while the other is shaded by a building. Reporting one number for a whole district therefore hides the heat that people are actually exposed to as they walk through it.
How it developed
2025 Summer
Jun 2025 – Aug 2025- Predicted localized temperature exposure across different times and locations using a Sky-View Factor approach — the share of the sky hemisphere a location can see, which determines how much of the sun's path is blocked.
- Performed GIS work on multi-source geospatial data (3D building morphology, building shadows) and microclimate parameters (solar radiation, temperature) using QGIS, CSDI portal, Open3Dhk, and Overpass Turbo.
2025 Fall
Sep 2025 – Dec 2025- Developed an urban building shadow analysis framework with seamless map merging for dense urban environments.
- Investigated sunlight pattern determination and visualization at minute-level intervals with high temporal resolution.
- Investigated map merging strategies without boundary artifacts, so separately measured tiles join into one continuous map.
- Performed GIS work on building shadows of different morphologies, solar irradiance, and Hong Kong map extraction.
2026 Spring
Jan 2026 – May 2026- Developed a 24-hour building shadow system across the HKUST campus and expanded the domain to the whole Hong Kong region.
- Optimized building shadow computation using multi-threading and multi-processing techniques.
2026 Summer
Jun 2026 – Aug 2026- Developed a Python script for automatic geo-referencing of thermal images (DJI thermal JPG to geo-coded raster).
- Operated drones to capture DJI thermal JPG images.
2026 Fall
Sep 2026 – Present- Estimate spatial temperature across the whole Hong Kong region.
- Use various neural network models for land cover classification.
- Plan and operate drones to capture DJI thermal JPG images in specified regions.
Areas
Cathay Hackathon 2025
Sept 2025 – Oct 2025 · participant
- Proposed a RAG-based AI chatbot for airline transit information, converting UX challenges into technical proposals under time constraints.
- Applied design thinking and collaborative brainstorming to improve passenger services.
Elsewhere
The full CV lists my degrees, coursework, awards, internships and tutoring roles in one place. For anything else, email is the best way to reach me.