AgentCon Hong Kong 2026, Hong Kong Institute of Information Technology, Hong Kong, Hong Kong
[Slides available] The workshop treats LLM-generated code as untrusted by default. Instead of relying on prompt scanning, it assumes the agent can and will fail, then uses infrastructure controls to contain the blast radius.
vLLM Hong Kong Meetup, The Hong Kong Polytechnic University, Hong Kong, Hong Kong
[Slides available] As large language models (LLMs) transition from research prototypes to production systems, inference becomes a primary bottleneck, and the gap between “it runs” and “it serves” widens quickly.
Hong Kong Python User Group Workshop, Auki Labs, Hong Kong, Hong Kong
[Slides available] This workshop introduces the Model Context Protocol (MCP), an new open-source standard for connecting large language models (LLMs) with external tools in a uniform, efficient way. MCP brings a modular, standardized approach to agentic workflows: LLMs can call any MCP “server” just as easily as importing a Python module.
PyCon Hong Kong 2024, City University of Hong Kong, Hong Kong
[Slides available] With the rapid advancement in deep learning, models become super large and consume significant resources, making efficiency and simplicity more critical than ever. In this talk, we introduce PyTorch Lightning, a deep learning framework that emerges as a powerful tool that streamlines the process of building, training, and scaling models, allowing researchers and practitioners to focus on what truly matters: innovation.
Hong Kong Python User Group Meetup, Oursky Limited, Hong Kong
[Slides available] With the growing complexity of Python projects and dependencies, managing them efficiently has become more crucial than ever. In this talk, we introduce Poetry, a modern dependency management tool that simplifies and enhances the process of handling Python dependencies, enabling developers to focus on building scalable and maintainable software.
2024 American Physical Society April Meeting, SAFE Credit Union Convention Center, United States
According to stellar formation and evolution models, black holes in the mass range of about 60-130 solar masses cannot form from stellar collapse. However, gravitational-wave detections, such as GW190521, display component black holes within this instability gap, challenging traditional models of stellar-origin black hole formation. Plausible scenarios, like hierarchical mergers in dense environments such as stellar clusters and active galactic nuclei, emerge to explain the presence of these high-mass black holes. Such scenarios, however, require the remnant black hole (later observed by our detectors) to inherit a gravitational recoil smaller than the escape velocity of the host environment.
48th International Exhibition of Inventions Geneva, Palexpo, Switzerland
There are thousands of skyscrapers can be found around the world and over 1300 of which in Hong Kong are standing over 100 m tall. Lift is in no doubt an essential mode of transportation in our daily life. Here we note the traditional way of lift doors inspection is unsafe, tedious and time consuming which cause long service outage. We, therefore, proposed the solution of Intelligent Lift Door Inspector, which is specifically designed for lift doors inspection to enhance the human safety at work and increase the availability of lifts.
Annual Physics Student Conference 2020, The Chinese University of Hong Kong, Hong Kong
When a gravitational wave passes by a massive object, its trajectory gets curved which can result in multiple images separated by a time delay, a phenomenon known as strong lensing. In the study, we propose three statistics to analyze the overlap of two sky localization probability maps, which could provide a quick preliminary lensing analysis of two gravitational-wave signals within 3 seconds.