Re: [Article] LLM Bot that reviews NuttX Pull Requests
Yikes Sebastien On 9/29/24 00:27, Lee, Lup Yuen wrote: We're experimenting with an LLM Bot (Large Language Model) that will review NuttX Pull Requests. This article explains how we created the LLM Bot in One Week: (1) We call GitHub API to fetch NuttX Pull Requests (2) Append the PR Body to the NuttX PR Requirements (3) Which becomes the LLM Prompt that we send to Gemini API (4) Our Bot posts the Gemini Response as a PR Review Comment (5) Due to quirks in Gemini API: We use Emoji Reactions to limit the API Calls (6) Though our LLM Bot was created by accident. (It wasn't meant to be an AI Project!) Please check out the article: https://lupyuen.codeberg.page/articles/llm.html Lup
Re: [Article] LLM Bot that reviews NuttX Pull Requests
On Sun, Sep 29, 2024 at 9:16 AM Simon Filgis wrote: > Hi Mr. Lup, > > I like! > > The LLM should not only check the commenting around the code. > > It should also try to review the code. > 1. Commented and easy to understand? Simplify naming suggestions. > 2. Efficient algorithms? Suggest different implementation, for example > better sorting algorithm. These ^^^ could possibly benefit from LLM, while... 3. Math errors (brackets, casting, rounding, overflow) prevention. > 4. Wrong or missing check of returns detection. These ^^^ could benefit from traditional static analysis tools. I think there are even commercial static analysis tools that provide free checking for open source projects. Perhaps something like that can be hooked up to PRs somehow and automatically post a PR review comment? 5. I guess many more! I definitely like the idea of using technology to assist in PR reviews. Cheers, Nathan
Re: [Article] LLM Bot that reviews NuttX Pull Requests
Hi Mr. Lup, I like! The LLM should not only check the commenting around the code. It should also try to review the code. 1. Commented and easy to understand? Simplify naming suggestions. 2. Efficient algorithms? Suggest different implementation, for example better sorting algorithm. 3. Math errors (brackets, casting, rounding, overflow) prevention. 4. Wrong or missing check of returns detection. 5. I guess many more! What do you think? Regards, Simon Venkat schrieb am So., 29. Sept. 2024, 05:28: > Interesting use case for LLM - Nice. Yes, a 50km hike can make such use > cases to be thought of! > > Venkat > > On Sat, Sep 28, 2024 at 3:27 PM Lee, Lup Yuen wrote: > > > We're experimenting with an LLM Bot (Large Language Model) that will > review > > NuttX Pull Requests. This article explains how we created the LLM Bot in > > One Week: > > > > (1) We call GitHub API to fetch NuttX Pull Requests > > (2) Append the PR Body to the NuttX PR Requirements > > (3) Which becomes the LLM Prompt that we send to Gemini API > > (4) Our Bot posts the Gemini Response as a PR Review Comment > > (5) Due to quirks in Gemini API: We use Emoji Reactions to limit the API > > Calls > > (6) Though our LLM Bot was created by accident. (It wasn't meant to be an > > AI Project!) > > > > Please check out the article: > > https://lupyuen.codeberg.page/articles/llm.html > > > > Lup > > >
