Back to News

Show HN: A CUDA-Generating AI Agent

Hacker News - AI
Jul 9, 2025 23:02
atallahw
1 views
hackernewsaidiscussion

Summary

MakoGenerate is an AI agent that automatically converts PyTorch code into optimized CUDA kernels, iteratively improving performance and correcting errors based on feedback from compilation and validation. Users can customize the process with prompts, documentation, and specific optimization goals, making it a flexible tool for accelerating deep learning workflows. This approach has significant implications for automating and streamlining the development of high-performance AI models on GPU hardware.

MakoGenerate is an AI agent that takes any PyTorch code as input, and writes CUDA kernels for it. You can choose a sample problem or enter in your own PyTorch. It generates CUDA, compiles and validates the results, then uses all of this as feedback into the next generation attempt. If the generated kernel didn’t work, it tries to fix it. If it did work, it tries to make it faster. You can configure the agent with a model/hardware/compute budget. The optional custom prompt section gives you the opportunity to add more context to the agent. You can include specific documentation, tutorials, or examples. You can also identify specific optimization techniques that you would like to see applied. Prompts can be as short as one line, or even multiple pages long. Let me know what you think. Comments URL: https://news.ycombinator.com/item?id=44515598 Points: 2 # Comments: 0
Read Full ArticleMore News

Related Articles

We Need AI Systems That Can Govern Themselves

Hacker News - AIJul 10

The article argues that as AI systems become more complex and autonomous, they must be designed with mechanisms to govern themselves, ensuring safety, accountability, and alignment with human values. It emphasizes the need for self-regulation and internal checks within AI to prevent unintended consequences, highlighting this as a critical direction for future AI research and development.

Every AI Integration Is Held Together with Parsing Logic and Prayer

Hacker News - AIJul 10

The article highlights that many current AI integrations rely heavily on fragile parsing logic, making them prone to errors and instability. This reliance on ad-hoc solutions underscores the need for more robust and reliable approaches as AI systems become more widely adopted. The piece suggests that improving integration methods is crucial for the long-term success and scalability of AI applications.

Show HN: I Built Fluent – macOS AI Writing Assistant

Hacker News - AIJul 10

A developer has launched Fluent, a native macOS AI writing assistant that offers instant text translation and improvement across any app via a Spotlight-like smart panel. Fluent supports formatted content in Chrome-based apps, customizable shortcuts, and integration with both cloud and local AI models, all without a subscription model. This tool highlights growing demand for seamless, user-controlled AI writing assistants that prioritize privacy and flexibility.