[Talk Proposal] KReview: AI-Powered Patch Review for the Linux Kernel
S, Sanath
[AMD Official Use Only - AMD Internal Distribution Only]
Hi All,
I'd like to submit a lightning talk proposal (10+5 minutes) for the upcoming BLR Kernel Meetup.
Title: KReview: AI-Powered Patch Review for the Linux Kernel
Abstract:
If you've submitted patches to the Linux kernel, you know how the review process works — maintainers are thorough, and rightly so. But quite often, patches go through multiple rounds of review for issues that are fairly mechanical: a missing Signed-off-by, an incorrect commit message format, an unhandled error path, or a style mismatch with the rest of the subsystem. These are real issues, but they're also the kind of things that can be caught early, before the patch ever reaches the mailing list.
That's the idea behind KReview — a simple, self-hostable AI-powered patch review server. You deploy it on a server (Linux, Windows, or any machine with Python), and your team can start getting automated feedback on patches right away. What makes it different from tools like GitHub Copilot code review is that it doesn't require you to change your workflow at all. There's no need to create a pull request or move to a different platform. Developers submit patches the way they already do — via git send-email or a quick HTTP call — and get back kernel-style inline reviews within seconds.
Under the hood, KReview runs a multi-pass review pipeline. The first pass identifies potential issues, the second filters out false positives using full source context from a local kernel clone, and the third uses a self-debate mechanism where the LLM argues both sides — author and reviewer — before deciding what's worth flagging. The system also builds a knowledge base by mining subsystem-specific review patterns from lore.kernel.org, so it learns what maintainers in a particular area tend to care about. You can plug in any OpenAI-compatible LLM backend, and adding support for a new subsystem is as simple as dropping in a markdown file.
In this talk, we'll walk through how KReview works, show a live demo, and share some lessons learned from building and using it. The goal isn't to replace human reviewers — it's to help developers catch the low-hanging fruit early so that maintainer time is spent on the things that truly need human judgement.
Outline:
- The problem
- KReview architecture
- Live demo
- Lessons learned and what's next ?
Speaker Bio:
Sanath S
I am a Senior Linux kernel developer at AMD, spearheading upstream development across USB4/Thunderbolt, NTB, xHCI, and DMA subsystems. I am the go-to engineer within AMD for all NTB and USB-related upstream activities, and with deep expertise across AMD's platform driver ecosystem — including PMF, I2C, I3C, SFH, UART, and SPI — I own first-level triage of all AMD platform driver issues surfaced by the community. I own feature enablement across all of AMD's client and EPYC platforms, ensuring that every platform driver is enabled and functioning seamlessly from post-silicon through upstream acceptance. Outside of my core subsystem work, I am actively exploring how AI can be leveraged to transform Linux kernel development workflows.
Co-author: Shyam Sundar S K
Looking forward to the meetup!
Thanks,
Sanath S
S, Sanath
Balakumaran Kannan
--
You received this message because you are subscribed to the Google Groups "Kernel Meetup Bangalore" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kernel-meetup-ban...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/kernel-meetup-bangalore/MW4PR12MB701669D5BCCC301D25983A548154A%40MW4PR12MB7016.namprd12.prod.outlook.com.