
we optimized inference for our table model and made it 4x faster on a single H100, median latency on our Light table model dropped from 3.35s to 0.76s our ML engineer @joon_kim96 wrote up our learnings and how we ultimately ended up using DFlash:


co-founder @ExtendHQ

we optimized inference for our table model and made it 4x faster on a single H100, median latency on our Light table model dropped from 3.35s to 0.76s our ML engineer @joon_kim96 wrote up our learnings and how we ultimately ended up using DFlash:


The example for this that I always give people is ordering. At every order counter there is a good old fashioned natural general intelligence (the employee) standing in front of you and yet you still prefer to scan a designed menu, rather than ask the employee a 100 questions
Interfaces are not going away. Some people keep saying the future is pure chat. "No UI, just talk to it." If that were true: - no meeting would have a whiteboard - pitch decks wouldn’t exist - You wouldn’t look at your phone screen You always need to show things and interact visually: drawing lines, previewing, whatever. Those needs don’t disappear because AI got better.

Keep your agents safe from document fraud with Extend Verify Also check out this great video of @kushalbyatnal, I unfortunately didn't make the cut this time after my abysmal performance on camera in our Series A video
AI-generated fake documents grew 500% last year, and humans can no longer tell the difference That's why we’re launching Extend Verify, the first document fraud detection API built for agents Verify catches the fastest growing types of document fraud, including: - fully AI-generated documents - localized AI edits (down to a single modified character) - patched text, overlays, and in-painting from Photoshop, PDF editors, and coding agents - digital tampering after a document was signed - and many more In our internal evals, Verify detected over 96% of AI-generated and tampered documents. Verify is available today: extend.ai/verify/request-acc…

This feels even more relevant now with everyone using Instinct to shop online Was just a bit too early
Thinking about the upcoming era when every store has an embedded AI sales agent, but every customer has a personal AI assistant that crawls stores for them to find deals. I like to imagine the agents will just have brief chat before getting down to business.

Few would catch this, but by design the drink strength on this menu corresponds directly to how hard the document is to extract from
best part about moving into a new office is getting to throw a banger office warming party on the Brex card Extend NYC 08.27.26




how it started vs. how it's going @opendoor chose Extend to power document processing for its title and escrow operations 10s of millions of pages, 300k+ homeowners, and we're just getting started
If you are a great engineer and have experience with AI document processing and OCR, please send me a DM - especially if you are in or are willing to relocate to Miami.

Our customers so often ask us how to recreate the document oriented UIs in our Studio, and we've never had a good answer for them until now Excited to finally open source some of our work here so anyone can make beautiful experiences for their document agents @andrewlu0 crushed it on this, go check it out!
Introducing Extend UI — open-source components for document agents - 14 components & examples for PDF, DOCX, and XLSX viewers, plus bounding box citations, file upload, e-signature, and more - fully customizable - MIT licensed when we started, we tried every file viewer and document component library we could find unfortunately, none of them had all the functionality (and polish) that we wanted, so we ended up building our own for @ExtendHQ it was only ever meant to be internal, but enough customers kept asking for it that we decided to give it back to the community it's useful for building agents, user-facing document flows, or internal tools we use and maintain it for Extend ourselves, so it'll keep getting better over time (and it's battle tested on millions of pages running through our system every day) it also works with design system agents like @magicpatterns for faster exploration and prototyping available today on the @shadcn component registry! some examples in 🧵


Long array extraction is a core capability we have invested a lot of time in even since the early days at Extend. It's a very challenging problem that is far from just a model problem, you need a purpose built harness that enables foundation models to reliably extract 1,000s of data points over hundred to get the most out of any top model. Our MAX mode for extraction does exactly that through a combination of things like: - Dynamic chunking of large documents based on table sizes/density and schema complexity, with semantic preservation as much as possible - Multiple passes through the full document to make sure all split context is persisted across the extraction over a long document - Heavy usage of smaller models used to detect and fix mechanical issues around a variety of page and section boundary conditions All this together brings us closer to the end goal of *perfect* extraction over any sized document and schema complexity. The most exciting part though is this is using a system we built and launched months ago, we're now working on a v2 that will take it to another level of complexity handling, stay tuned 👀
we created a new, open source eval (LongArray-Extract) for one of the hardest problems in document processing: how to extract every row out of long documents some highlights: - Extend's array extraction is SOTA (99.2%) - 3x faster than the next closest competitor (5 min vs 14 min) it's based on examples we've seen in production: > bank statements with 2,000+ transactions > clinical adverse-event listings with 1,000+ events > legal filings with hundreds of numbered factual paragraphs if you've ever built a document pipeline on hundred page docs with thousands of listings, you know exactly how quickly things break we open sourced the benchmark + dataset so teams can inspect the docs, run the harness, and compare results directly

Few understand this, but there is no better time to reflect on how to improve your document ingestion pipelines than on your morning commute in nyc
why are there AI ads on the nyc subway now... how do i get out of this bubble designed by @AirfoilStudio ????


There are many things I love about our new site, but my favorite is the animation all the way at the bottom for those interested enough to scroll to the end
proud to share Extend's updated brand and website! we spent 100s of hours on it, and obsessed over every single detail multiple full redesigns thrown out, fonts swapped and swapped again, every animation tweaked until it felt right...we even locked ourselves in a room and white-boarded every single word on the page until it resonated why? prospects would see a demo and say something like "this is not what I expected based on your site", and they were 100% correct our product has changed so much in the past year, that we wouldn't even recognize the old version (new APIs, capabilities, entirely new categories of problems we now solve). Our old site didn't reflect any of that. we’re proud of how this turned out, and we hope it conveys the level of craft our team obsesses over in everything we ship check out the new site below and please share feedback!
