News for hackers. Stuff that matters.

ykjs

Re:Ask HN: Who is hiring? (February 2026) on Thursday February 12, 2026 @05:18PM (#46996070)
JigsawStack | Founding GTM (Go to market) / Growth | San Francisco, London, India, REMOTE (US) | Full Time

Company site: https://jigsawstack.com

At JigsawStack, we’re building specialized small models that automate complex backend tasks like OCR, web scraping, classification, data extraction, and more. We power thousands of developers and process billions of tokens every month.

We’re hiring our first Founding GTM, someone who wants to architect the entire GTM engine, experiment aggressively, and push the boundaries of AI growth.

You’ll own automation, outbound systems, GTM tooling, growth experiments, and the workflows that expand JigsawStack’s reach. This is a growth role where you’ll build pipelines, write scripts, automate everything possible, and systematically scale distribution.

As an early team member, you’ll work directly with the founders, run high-velocity experiments, and shape JigsawStack’s playbook for years to come.

More info and apply here: https://yoeven.notion.site/founding-gtm

Email here for more questions: [email protected]

Show HN: I built an open source LLM integration for PostgreSQL

https://github.com/JigsawStack/postgres-llm

More often than not we pre-process data using an LLM before inserting them into a column/row on the database, for example translating text and storing the different languages or running OCR on a document and storing the structured output or using Classification to label user messages/documents before storing it.Many of this use cases could happen on the database rather than a dedicated...

( Read More... | no comments | on HN )
Hey! Awesome product, not sure which model you use under the hood but you should check out https://jigsawstack.com/docs/api-reference/ai/vocr. Great for ID extraction which also includes bounding boxes and other attributes you'd need

Show HN: Interfaze – The LLM Built for Developers

https://interfaze.ai

LLMs have changed the way we code, build, and launch a product. Many of these cases are human-in-the-loop tasks like vibe coding or workflows that have a larger margin of error that is acceptable.However, LLMs aren't great for backend developer tasks that have no/low human in the loop, like OCR for KYC or web scraping structured data consistently or classification. Doing all this at scale and...

( Read More... | 1 comment | 3 points | on HN )
Awesome stuff! This really helps
Can this be provided as an API?
It doesn't, it's a google translate widget alternative which is Open Source. Google deprecated their translate widget hence you can't really use google translate for this use case. So it's an open source framework that uses another translation service that can support this widget on the front end with the right security controls to prevent spam, etc.

The first comment explains the full story.

Hey HN! The easiest and fastest way to get a localized website was to add a single line for the Google Translate Widget script in your head tag.

Unfortunately, the Google widget was deprecated and has never been updated to match the modern standards of sites like React frameworks and many more. While the translation quality, especially, has lagged compared to where we are with AI today.

So I built the a open source translation widget that you can do a quick one script add and translate your website to 100+ languages. Comes with other fancy features like local storage caching, react page change watches, etc so you get the full package out of the box.

Re:Show HN: Image-to-Image Translation Model on Monday June 02, 2025 @06:06PM (#44163648)
Yes next version will have over 98 languages :)

Show HN: Image-to-Image Translation Model

https://jigsawstack.com/blog/translate-any-image

We launched a v1 of a image to image translation API which translates the text on an images by replacing the existing text.For v1, it's pretty much a model pipeline: OCR current text -> generate mask -> erase text -> translate text -> use embedding comparison to find similar font -> map text back on imagev1 was more like a prototype which already beats many of the similar...

( Read More... | 2 comments | on HN )
Re:Vibe coders don't know what they're paying for on Wednesday May 28, 2025 @12:40AM (#44112797)
We have a customer who has zero technical background and was vibe coding using Cursor. Basically hitting "Accept" all the way. At some point, Cursor suggested to install the JigsawStack SDK which then eventually prompted for an API key.

Naturally, he put his credit card down and got the API key on a $27/mth Pro Plan which includes 8m tokens of usage every month followed by $1.40 per million tokens.

A week later, he's on our support email surprised by a $200+ charge on his card. He was pissed, angry demanding a refund. Typically for situations like this, we do a partial refund to cover the cost for first-time customers if they honestly made a mistake and used more than intended.

That's when we realized, we didn't have a single $200+ charge tied to his account, only the $27/mth charge and we were so confused. We asked for the bank record/statement/invoice ID or anything that can help us find this charge.

That's when he sent an attached Cursor support email and invoices! And then all the pieces came together. He thought Cursor and JigsawStack were the same company because Cursor suggested to install the JigsawStack SDK. He got a $200+ charge from Vibe coding too hard! He was using max mode which is like 0.05 a prompt or something around those lines.

As you can see we have a language barrier as well. We tried our best to explain how these are two different companies and we weren't the ones who charged him. I think he got it since he's still a paying customer :)

Re:O1 Visa Question - Looking for help on Monday May 26, 2025 @07:29PM (#44102593)
Hey hey! Membership and Judging are the easiest items to get, and that's why i think your attorney really wants you hit that.

You don't have to join a accelerator where you have to give up equity. You could do something like ODF, or any program that has a selection process, is somewhat known and has a acceptance rate criteria. You could frame your VC as a membership too. For example YC is both a membership and a VC, Antler another. As long as your VC has a small membership page or program somewhere, you could frame it as such.

Judging is super simple, reach out to hackathons hosted by your friends, or in your area and ask if you can judge. Do a few, maybe 2-3. Try to get competitions where you co-judge are from companies with brand names like AWS, google, etc. That helps too.