CUDA matrix multiplication benchmarking on Jetson Orin Nano. Four implementations, three power modes, five matrix sizes. 99.5% mathematical validation. C++/CUDA and Python.

FlashAttention-style CUDA implementation with shared-memory tiling, online softmax fusion, IO-aware optimization, and GPU benchmarking.

DeepSeek-R1 7B INT4 at 69.3 tok/s on a $300 RTX 3060. Faster than llama.cpp, vLLM, and NVIDIA TensorRT-LLM. Is one developer + Ai really better than the entire industry?

A 110M-parameter Llama-style transformer trained from scratch on the TinyStories dataset, optimized for high-throughput training on 4GB VRAM consumer GPUs. The project features a custom asynchronous CUDA-stream prefetcher and KV-cache inference, achieving 10k+ TPS on an RTX 3050.

🏆 Which 3DGS renderer is fastest? Which compression is best? We measured them all — on the same GPU, same scenes, same protocol.

High-performance matrix engine for Unit-Domain Flow (UDF). Eliminates Mantissa Friction with 0.00 MSE integrity.

Hardened RAG pipeline with Llama 3.2 (3B) & Arize Phoenix. Features 4-bit Unsloth optimization, OpenTelemetry auditing, and a KV-cache stability patch for T4 GPUs. P99 Latency: 19.2s.