25-26 Third T19

T19 Session is the main session of GDGoC Yonsei.

• In the APT Update Session, speakers present the latest technologies and emerging tech trends.
• In the Tech Session, presenters share on topics of their personal interest. This time, alumni also joined as speakers.

✔️ APT Update

Cloud Member – Seokwon Lee

Introduced MCP, a protocol that connects LLMs with external programs through a standard interface and returns tool execution results as context. He demonstrated projects such as React integration with Context7, screen flow design with Figma, browser automation using Playwright, and scheduling with Task Master. At the same time, he pointed out several security vulnerabilities of MCP.

✔️ Tech Session

1st Gen Front-End Lead – Minseo Kang

Outlined the four core aspects of UX for front-end: speed, interactivity, stability, and accessibility. Speed: SSR & Hydration, code splitting, image optimization, caching, preload (for fast initial rendering); prefetch and CSR (for instant navigation). Interactivity: throttle & debounce, React concurrent model (Fiber), Optimistic UI. Stability: font-swap, image aspect-ratio, placeholder, and transitions to ensure predictable UI. Accessibility: localization (L10N), tab index, and image alt text to ensure inclusive usage.

1st Gen Lead – Giwoong Jeong

Explained PyTorch’s eager mode (immediate execution) and graph mode (optimized execution through computation graphs). He introduced modern DNN workloads where manual dispatch coexists with compiler-generated methods, and how stacks like TVM accelerate models such as ResNet through kernel fusion and scheduling optimization. He also emphasized how DSLs like Triton and CuTe allow developers to write high-performance GPU kernels concisely in Python syntax, achieving both performance and productivity.

2nd Gen Back-End Lead – Hoesung Yoo

Presented best practices for server architecture. Starting with separating web servers and application servers for different workloads, he explained that traffic surges are generally addressed through scale-out and that load balancers handle routing, health checks, SSL termination, and balancing across servers. For reliability, removing Single Points of Failure (SPOF) is key—achieved by configuring active-active or standby load balancers. He also discussed solutions for databases, such as locks for single DBs and read replicas for scaling.