Coding Projects

Recipe for running Qwen3.8-Flash-Next (~176B params, 6B active) on a single NVIDIA DGX Spark with vLLM, using Intel's W4A16 AutoRound int4 checkpoint plus int8 lm_head and fp8 side layers, and an fp8 n-gram PLE table served from local NVMe via mmap, or from another machine's RAM over one-sided RDMA READ. Achieves ~49 tok/s decode with MTP=3 speculative decoding, working prefix caching, and never-evict KV pinning for system prompts. Fork of blazux/qwen3.8-Flash-DGX.

WTAKO Network's Frontend (this website) is written in Next.js 16 and MobX, with the help of a local LLM in my AI core inside my home. I just hope it looks good enough because I don't have much design sense. This Next.js project is exported as a static site and it is being hosted on my own server. I never liked the idea of Server-Side Rendering (SSR), and I also avoid Vercel as much as possible.

Hardware & IoT Monitoring Server with real-time Linux hardware monitoring and IoT device tracking via WebSocket. Collects CPU, memory, temperatures, fan speeds, storage health, and network statistics.

Cantonese-speaking LLM voice assistant for Home Assistant with natural conversation, tool calling for KMB/HKO APIs, and AI radar analysis. Features ~0.15s STT latency and persistent memory via HA todo. See it in action: https://www.threads.com/@nekomata.saren/post/DXV4Z-AkXp4

Real-time Speech-to-Text system optimized for Cantonese with hotkey-triggered recording, VAD, and automatic typing. Supports GLM-ASR, SenseVoice, Whisper, and Qwen ASR backends. Includes SRT generation/translate workflow.

Chinese handwriting input method exploiting Google Translate via Electron. Features macOS-style touchpad support for gesture-based character input. Designed for Linux X11 desktops.

Android forum reader app for browsing and participating in online discussions.