tanglearncode
Joined 4 months ago
Karma
5
Submissions
16
Member since
Jul 11
Recent submissions
Comment · 3 months ago
Comment · 3 months ago
A claude code plugin is now open sourced: https://github.com/XTSoftwareLabs/neatcontext-plugins
It extracts domain knowledge and save useful work from a Claude Code conversation as structured, reusable context that you can use in later sessions or share with others.
Comment · 3 months ago
Hi,
I recently researched the hard problem for the inaccurate answers from SRE agent. I have a different opinion for the SRE agent in the market. TBH, what an oncall engineer only needs context +AI client to handle incidents. I don't see how SRE agent can be better. The key is how to organize the context, not the AI.
So I wrote the blog in the link to explain it in details. Happy to discuss more.
Comment · 3 months ago
I just made a video to show how the domain knowledge issue can be solved: https://youtu.be/lh5ZgP6CHss
Comment · 3 months ago
A little more context on this. I have tried quite a few of SRE agents or AI incident handling assistants in the market. LLMs can help on incident handling if the domain knowledge and context are clear. However, in most of time, they are missed. SRE agents usually are belong to a platform. Naturally the platrom companies want users to keep within them. However, the domain knowledge are anywhere inside user's company. Every team has its own domain knowledge to impact how to handle an incidents in real time. This results inaccurate response from AI.
I'm trying to solve this problem in another approach. Create a lightweight desktop app, let the user connect whatever domain knowledge they have, whether a doc inside company, or runbooks, or TSGs or even some rules within a team only. Let the user connect to their own internal systems. Then provide those context to LLM. It works pretty well.
Here is the hands on demo to show exactly how it works: https://github.com/XTSoftwareLabs/neatcontext-demo
Comment · 3 months ago
Yeah the context is the key to make the LLM really useful for oncall. But honestly the SRE agent companies today are trying to keep users in their platforms, not focusing on solving the context problem. That's what I saw.
Comment · 4 months ago
Some lessons learned when using the SRE agents to handle incidents. We eventually ended up to a semi automation way to get more accurate LLM results by providing domain knowledge and context to LLM. Wondering if you encountered the similar inaccurate or sometimes nonsense response from SRE agents. How would you solve it?
Turn conversations to structured domain knowledge. Across sessions. Across AI agents. The stored domain knowledge is human readable and can easily shared with others.