CUDA Army

Name: CUDA Army
Author: Luis Guerra

For ML engineers: CUDA Army optimizes your neural networks on Nvidia GPUs so training and inference run faster with custom kernel tuning.

Built by Luis Guerra · Free

Visit website

Instant Launch listing

Listing status

9upvotes

Community score

Luis Guerra (Joined May 2026)

Founder

Compared against 6 alternatives

Market comparison

Launched on Launch Llama May 30, 2026 · #9 of 17 that day

About

We optimize you neural network training and inference pipelines for your target hardware.

We are specialized in Nvidia GPUs and libraries: CUDA, CuBLAS, CuTLASS, CuDNN, CuTe, NCCL, NVSHMEM. However, we can also work on other platforms and DSLs on-request.

We are an experienced team in quantization, pruning, distillation, distributed systems, parallelization, sharding, compilers and custom kernels such as Flash-Attention variants.

We also train custom models using your proprietary data while maintaining privacy and security. We have done projects on the following tasks:
•Computer Vision & Image Processing
•SLAM & Robotics Systems
•Reinforcement Learning
•Large Language Models & Fine-tuning
•3D Graphics & Rendering

Use Cases

Speeding up neural network inference by writing custom CUDA kernels tuned to a specific model and NVIDIA hardware
Scaling AI training and inference across multiple GPUs or multiple nodes in large deployments
Reducing memory use and latency through INT8/FP16 quantization and Flash Attention optimization of transformer and LLM workloads

Who it is for

ML engineers optimizing neural network training and inference on Nvidia GPUs.
Enterprise AI teams moving prototype models into production on their own infrastructure.
Teams building computer vision, robotics SLAM, reinforcement learning, or LLM fine-tuning systems that need proprietary-data model training.