A lightweight, high-performance robotics communication framework. Serves as a zero-bloat alternative to ROS2, featuring zero-overhead serialization and deterministic real-time communication.
A deep learning library written from scratch. Includes hand-optimized CUDA kernels, a custom graph-based scheduler, and a minimal JIT backend to eliminate high-level runtime overhead.
An intelligent API security engine. Features a zero-allocation high-concurrency ingestion pipeline coupled with a low-latency native detection core designed to parse traffic at line rate.
A pre-print research paper on multimodal fusion layers in transformer architectures, exploring linear-scaling attention mechanisms for vision-language models.