ML Systems & Compilers Yoshida Production-grade General LLM Reinforcement Learning & Post-Training Framework NInfer (RTX 5060 Ti) High-performance C++/CUDA single-GPU inference engine for NVIDIA Blackwell architecture (RTX 5090 & RTX 5060 Ti 16GB) AuK Optimization (RTX 5060 Ti) Bare-metal inference & ultra-low-latency streaming engine for Tencent Hunyuan AuK-Flash 1.5B on NVIDIA Blackwell SM120 Axon A minimal deep learning framework built in C++20 from scratch with eager autograd, AVX2 SIMD CPU backend, and GGML-style quantization Flash-SDXL High-throughput end-to-end web serving engine and inference optimization pipeline for SDXL models Angel Next-generation distributed LLM RL framework with VERL-style 3D HybridEngine architecture (In Development) Generative Models & Foundation Systems Project Yoru & Fami Stealth research on next-generation image generation and speech/audio foundation systems Batik MeanFlow Generative Flow Matching and Mean Flow applied to manifold learning and Indonesian Batik pattern synthesis Waifu Diffusion Exploration Fine-tuning and latent diffusion conditioning experiments on stylized generative image models Bernini MNIST Generative flow and velocity field modeling experiments on benchmark vision datasets Media Infrastructure & Dataset Tooling Booru Ecosystem & Dataset Harvesting Suite Unified dataset ingestion, smart tracking, automated tagging, and scheduling pipeline for mediaboard resources Full-Stack & Interactive Applications InSpaceMap Platform Full-stack indoor venue mapping, drag-and-drop navigation graph editor, and wayfinding platform Tebak Kata Contexto-inspired semantic word-guessing game built with Next.js and vector embedding similarity matching IdleRPG Engine Automated RPG game engine, progression state machine, and battle mechanics backend
