ArchTA
Part-time Research Assistant, Team member, & Researcher
Sep 2025 – Jun 2026
This project plans to train a general multi-modal model for architecture, localize it for HKU, and build AI tools for formative assessment and gamified learning.Keywords: LLM, AWS, React
Under what tasks and conditions can an AI teaching assistant reliably improve teacher orchestration and student iteration?
Can AI-derived progress signals increase the effectiveness and fairness of gamified interventions?
Build a maintainable, extensible, modular, offline-first architecture that supports easy A/B variant switching (e.g., with/without gamification).
How to enable one-stop deployment, modular multi-LLM orchestration, integration with assignment analytics, dashboard analytics, and chatbots.
We plan to introduce gamified design elements (such as points and dashboards) to quantify each student’s positioning and progress along different personalized paths in architecture, thereby encouraging consistent learning. By incorporating RAG (Retrieval-Augmented Generation), the system will surface targeted reference cases to support students’ day-to-day study.Finally, leveraging the expert model’s interpretive capabilities, we aim to evolve from a basic architectural Q&A chatbot into a context-aware teaching assistant that can understand multimodal student work and learning states, and provide guidance accordingly.As the project, I will be responsible for designing and deploying the overall cloud architecture for the project, as well as participating in the training of large models. I will also participate in publishing two papers spanning education, HCI, and AI.
Next‑generation Exhibitor System
Senior Developer, IBM International Co., Ltd., HK
June 2025 - Present
Joined in full-cycle scalable cloud solutions, including design, framework setup, coding, and CI/CD deployment for the Exhibition+ 's projects.Keywords: Cloud, AWS, DDD, Full Stack
Adopt microservices, cloud services on Azure and AWS for infrastructure and modular application design
Leveraging StreamSets and Kafka for data streaming.
Full stack Technicals:Spring Boot, React, Express, Form Builder
Cloud services: AWS services including: CloudFormation, ECS, Lambda, IAM, API Gateway, etc.
Adopts the mature Domain‑Driven Design (DDD) approach, which models the system around the business domain.
Built the Foundation Transaction Log service using Debezium for CDC and Kafka for event distribution.
Deployed multiple AWS Lambda + SQS + API gateway services. components for event processing.
Found a solution (POC) to support PDF generation.
Added ECS logging to s3 for system's microservices.
Other Type.js and SpringBoot + Java development tasks.
Joined IBM as a Senior Developer on my second major project; leveraging full‑stack experience and independent problem‑solving skills. I independently built or investigated multiple services.The Exhibition+ system is a next‑generation platform for exhibitors to apply for booths. The system can support different types of local Hong Kong and overseas fairs, conferences, forums, and seminars. The promoted exhibition model combines online and offline elements, extending the face‑to‑face interactions and business negotiations of physical events into an online intelligent matching platform to connect business partners.It is a scalable cloud-native solutions.
SMS CDR BigData To DataLake Pipeline
Developer | System Operator, CITIC Telecom International, HK
December 2022 – April 2023
Contributed to parsing on the legacy SMS processing platform and developed the Process CDR File part of the new system.Keywords: StreamSets, Big Data, Data Lake, Kubernetes, telecom switch, CDR files
What is the impact of AI-augmented interaction on interaction fluency and perceived experience in creative design?
How does collaborating with AI reshape reasoning patterns, attention allocation, and exploration behaviors?
How do participants negotiate agency and trust with AI suggestions and scaffolding over time?
Build a fault-tolerant StreamSets pipeline to fetch SMS CDR files from SFTP, parse and split them into records, and trigger a completion event per file.
Route records by A2P/P2P, enrich and validate fields, then write valid records to Kafka/HDFS and invalid ones to an error store.
Log splitter audit info (file name, record count, status), support retries, monitoring, and replay for fault-tolerance.
Project Achievement: Enhanced SMS system performance by reducing process time by 30%.
The project build a StreamSets pipeline to ingest SMS CDRs from SFTP, parse and send valid records to Kafka, HDFS, or error store.
VRGaze 2 AIGC
Concept Designer | Main developer, Hobby project
Nov 2024 – Dec 2024
The project analyzes visual fixation patterns while users view images in VR to infer their preferences, then leverages these inferred preferences to generate new images, creating a closed-loop iterative process.Unity (Oculus), ComfyUI, OpenAI Version
How can AI infer user preferences and proclivities from attention signals, and leverage these in a continual, user-centric generative pipeline (AIGC)?
Applied OpenAI Vision to analyze user interests based on gaze data.
Leveraged ComfyUI to generate and iteratively refine personalized AIGC content.
Established a continuous feedback loop to enhance content personalization and user engagement.
The project built a Unity workflow to capture user attention and preferences through VR gaze tracking.
It integrate Meta's VR SDK into Unity to capture users' gaze in real time and detect gaze hotspots, take corresponding scene screenshots to send to GPT Vision for keyword extraction, and then based on those keywords generate or retrieve images, that is ultimately presented in VR.
Robotic Car
Team Leader | Embedded Developer, Bachelor’s Project
Sep 2022 - May 2022
STM32-based robotic cars with independent wheel PWM and PID control for precise line-following and stable motion, featuring wireless inter-robot communication for synchronized baton handoff and cooperative relay laps.Keywords: Microcontroller, STM32, PID, ROS
Hardware: STM32 microcontroller as main controller; dual motor drivers with PWM for independent left/right wheel control.
Control: Implemented closed-loop PID control for each wheel to achieve stable speed and accurate heading; tuned PID parameters to minimize wheel speed mismatch and trajectory drift.
Sensing: Line-following using multiple infrared reflectance sensors with analog/digital filtering and sensor fusion for robust detection on curves and intersections.
Communication: Designed wireless communication protocol between cars using Bluetooth; implemented error checking and simple handshaking.
Relay Coordination: Developed a cooperative relay algorithm allowing two cars to exchange baton/state while maintaining line tracking; includes start/stop synchronization, handoff zone detection.
Software: Real-time firmware on STM32 with timer-based PWM, encoder feedback processing, PID loops, and interrupt-driven comms; modular code architecture for maintainability.
Achieved A grade