Vatsal
@vatsal_manot
Building @PreternaturalAI (YC W24). Maintainer of @SwiftUIX.
Excited to share that I've been working on something — a personal generative search experience. It's built on something completely unique — a family of native, hardware-accelerated frameworks written in pure, high-performance Swift. Not one single line of Python. More below:
TLDR: security and file management paranoia - Full answer - When you install an update, mac isn't just swapping out a few files - it's doing a ton of behind-the-scenes verification that takes forever. - First, it's checking every single file's digital signature to make sure…
the next software giants will be built by those who systematically integrate domain expertise into AI systems, not those with the biggest models. come to our talk! we'll send you the recording and study notes if you cant make it live maven.com/p/504453/lesso…
Applied with less than a minute to go till the deadline. Cannot overstate the scope of the positive impact that it’s had on my life, W24 batch was truly a special experience for me.
Also did the 2 hour before thing, which forever altered the timeline of my life. If you have an idea, simply apply.
3rd, such a system by definition is a link/click washer. how do you filter for abuses? What do you tolerate? 4th, funding! Especially without angering target site 5th, stewardship. What liabilities do you have for malware that's being served via your system? 2
There's lots of logistical things. 1st, do you want to allow the link to be changed to point to a site if it's moved? If so, how do you control long term something that will need to last indefinitely? 2nd, what do you want to have happen if the resource moves? 1
sometimes idk how to process how much fake work gets counted as real work in this world!!!
i hope i’m not becoming bitter but im slowly despising the modern state of software more and more because software is so sloppy, badly designed and bug ridden that it feels insulting considering how much time it takes out my not-so unlimited time
IMO @chunkrai is to RAG what @rimelabs is to TTS. Both are extremely efficient and passionately competent at what they do, and are interested in furthering the contemporary rather than regurgitating the same slop over and over. Congrats to the Chunkr team on the launch!
Excels are notoriously hard to parse. Even though they are machine readable, they end up as a flat string that loses semantic meaning of tables, styling and formulas. Agents can’t use it. We’ve solved that. Raw spreadsheets to perfect HTML, live now on the @chunkrai API.
If you are a technical founder, you do not need a non-technical cofounder.
New blog post from Apple machine learning research on FastVLM: all about improving efficiency and accuracy. Plus there is an open source MLX implementation which runs on an iPhone:
🌟Explore key insights from the FastVLM project (real-time vision-language model) in this blog post: machinelearning.apple.com/research/fast-…
🌟Explore key insights from the FastVLM project (real-time vision-language model) in this blog post: machinelearning.apple.com/research/fast-…
A botched UI rushjob to make people stop talking about a botched AI rushjob. Some 200 IQ exec moves being made over there
My perspective is Apple spent three years working on a UI and shouldn’t throw it away over a week of Twitter posts. (Not that the UI is perfect). That’s how we ended up with an iPhone mini that nobody who said Apple should make one actually bought or would recommend.
on the rare occasion that I muse to myself, “hmm, Apple was right to keep this API private”
I like Little Snitch but it needs the @proxyman_app treatment, and Tiny Shield is just a tad too tiny. Anyone know of an OSS macOS app that is functionally equivalent to Little Snitch? Looking for a base I can hack atop.
Every time I try to seriously use SwiftUI for a complex new feature, I come away slightly disgusted with myself for having wasted so much time implementing the most basic things that there's millions of libraries for in React. Like, we're halfway to 2026 and you can still not…
It’s a dev lifecycle issue and/or Conway’s Law. In order to provide the LLM with full background and information on the whole product, you need to build a dev lifecycle that inherently represents those capabilities in a concise and up to date form at all times. I don’t think…
I can't even scream at the AI now? what am I even paying for