Vinay Soni

I’m an Independent AI Research Engineer focused on evidence-driven iteration for language modeling. Over 2+ years, I’ve designed and optimized language modeling pipelines—from tokenization (C++ BPE) and training data strategies to long-context mechanisms and efficient attention formulations—while benchmarking compute and storage trade-offs. In parallel, I build performance-first implementations with custom CUDA, Triton, and SIMD kernels, and I’ve explored hybrid context architectures and learnable quantization that dynamically allocates precision. Previously at Dell Technologies (7+ years), I delivered backend systems and product improvements, worked on a patent-related dictionary-less compression approach, and upgraded platforms for faster performance—bringing an engineering mindset grounded in measurable outcomes.