956 karma · joined June 17, 2014
https://twitter.com/helloteban https://github.com/baristaGeek https://www.linkedin.com/in/tebanvargas/
Feel free to ping me! estebanvargas94 [at] gmail [dot] com
[ my public key: https://keybase.io/baristageek; my proof: https://keybase.io/baristageek/sigs/xvrFueBx0GMNM1qPjertwxrGmgyPzGzabPoXl_Q02Io ]
Medellin, Colombia
I agree that SSR makes the "frontend VS backend API" separation very confusing. Session token VS secret key is a better framing for this, and there's a very interesting insight into the paradox that you described! The ergonomics you mention for ensuring consistency sounds like a very interesting technical challenge.
1. With the new server components paradigm, React and Next have gone through a lot of changes around how SSR is done. Are your helper functions (currentUser(), currentOrg(), etc.) compatible with both the old and new approaches? Or is adapting to both one of your current technical challenges?
2. Are you open to candidates in the Americas time zones (LatAm)?
Besides this, you can also see on our repository that we only send line diffs, titles of pull requests, and descriptions of pull requests to the GPT API, not your entire codebase.
That being said, we understand that many companies can't use OpenAI-powered products because of the lack of a data transfer agreement (ChatGPT Enterprise version aside).
Self-hosted tools aside, what vendors do you use that solve this problem well?
We're still discussing this. We do have a plan to open what's currently built entirely, and later build proprietary features inside an /ee dir.
You can read more about our strategy on our company's open handbook: https://github.com/watermelontools/public-handbook/blob/main...
From checking out their website, it seems like they've developed a more advanced way of chatting with line diffs, whereas we've built a superior PR summary system by tracing code context from various systems.
Currently, we exclusively integrate with OpenAI's models because they've been highly effective for our needs. However, we're definitely open to the idea of supporting open-source and self-hosted LLMs in the future. Thanks for bringing this up, and I hope this clears up any confusion!
I'll also be available to answer any questions from the community and to receive your feedback.
We still see this technology being able to handle edge cases (as you mention) such as certain movements of blocks of code to another file.
Personally, I'm more drawn to "applied AI engineering" – it's like a blend between data science and product engineering, encompassing areas like vector databases, prompt engineering, agents, chains, and so on. Meanwhile, I'm also picking up modern high-performance languages like Rust and Golang on the side.
Watermelon is your open-source copilot for code reviews. Our GitHub app dives into PRs, providing context using information pulled from platforms like GitHub, Slack, Linear, and Confluence, to name a few.
After seeing over 16k downloads of our VS Code extension, we're thrilled to launch its IntelliJ version. These IDE extensions serve as a front-line context provider for individual developers, while our GitHub app offers team-wide context. Think of it like GitLens, but with added insights from GitHub and other enterprise tools.
Would love to hear your thoughts and feedback!
GitHub Codespaces for quick stuff, I prefer my IDE.
I was just wondering if Spotify (a CEA) contributed to that slowdown.