Introducing interfaze-1.1-lite, our truly open-weight model for deterministic work
- 30% faster on document processing
- 44% less memory use on a single GPU
You can now combine and single shot tasks like:
OCR + Translation + Classification
https://t.co/WbOM7eqBFm
here’s a weekend gift for you (and your agents)
you can now run tasks with Interfaze without an API key or signup 🚫🔑
introducing Interfaze Keyless
extract text, parse documents, and turn unstructured data into structured outputs and more with just one API call
highest accuracy in the market (open source benchmarks on our site).
no setup. just start building for free.
60+ YC founders across batches from 2018 to incoming in 2027 showed up last night for our YC Generations gathering
So happy to see the builders community flourish and co host with:
@graypassinc (yc f26)
@Fondocom (yc w18)
@modaflows (yc w26)
with support from @JayChopra_
lots more to come from @interfaze_ai 🙂
Introducing Interfaze People Search 🔍
Go beyond names and job titles. Find researchers and inventors through their papers and patents, with exact source passages showing why they match.
Explore their work. Find email and social links where available. Connect with the people behind the research.
Open source. Powered entirely by @interfaze_ai
Introducing Interfaze Atlas 📄
Open-source tool to turn hundreds of messy PDFs into structured knowledge you can search, compare, and build on.
Group PDFs, extract key details, and export JSON with supporting quotes linked to source pages ↓
Things we do at @airwallex to support our founders. @yoeven started with us in SG and then transitioned to US. We also leverage his tech internally.
If you’re a founder expanding globally, swing by the office; there’s lots we can partner on 🏎️
holy shit just met a team burning $14,000/ month on frontier models while 3 of their engs still spend ~20h per week proofreading 200-page contracts
i'm sure you've felt this pain if you deal with any sort of document pipelines
which is why we built Interfaze with two key features: bounding boxes + confidence scores for each extraction
all at a fraction of the cost of LLMs
we want to give teams their time back to build things that actually matter.
Anthropic's entire IPO filing. 289 pages. 13 minutes. 🤯
@interfaze_ai pulled out all 21,760 lines and boxed every single one exactly where it sits on the page
try it for yourself for free: https://t.co/h7ND10ggC8
being the "best" at what you do will get you the meeting, but it won't close the deal
being #1 on OCRBench or OLM-OCR means nothing to us if our model struggles on our customers' actual production files
public benchmarks like these are built on a static dataset, but real-world documents and files are full of edge cases; this is why most people view public evals and leaderboards as just marketing.
Real documents are very messy, so at @interfaze_ai, we built a growing internal eval bucketed into real failure points like scanned files, tables, circular text, and weird formats
as we work with new customers, we keep adding to it so our models get better and better at real documents in production
every customer has their own private eval, whether they've written it down yet or not.
you win the customer when you pass it.
if you're applying to YC this winter as a technical founder, there's a brutal transition waiting for you.
good tech and architecture gets you your first few thousand $ of revenue.
scaling past that has very little to do with code, and everything to do with two pipelines you have to run nonstop:
1. Hiring
you need people earlier than you expect.
most technical founders wait until they're drowning in backlog, then make panic hires that burn months of runway to fix and are very meh
talk to talent and build pipeline before you have open roles.
2. Fundraising
when money is running out, every decision becomes panic-driven.
keep conversations going in the background so you never raise from a position of weakness
code gets you off the ground, but hiring and cash flow will keep you alive.
if you're applying for the W27 batch, reply with a one-liner about what you're building. if it sounds interesting, I'm happy to refer you.
The first 3 people to find me at @vercel Ship will get $5000 in @interfaze_ai credits
If you know just how cheap our model is, you know how long this will last you👀
Introducing interfaze-1.1-lite, our truly open-weight model for deterministic work
- 30% faster on document processing
- 44% less memory use on a single GPU
You can now combine and single shot tasks like:
OCR + Translation + Classification
https://t.co/WbOM7eqBFm
interfaze-1.1-lite is made up of many smaller weights like TimesFM 2.5, PP-DocLayout, SAM 2.1 and more
Each model is a specialist model focused on just one task
They can be automatically combined to build workflows and their results get merged into a consistent structured output
Thank you to all the amazing people building cool things who joined us at our #SFTechWeek Wellness Night!
We brought together CTOs and technical builders backed by some of the most prominent VCs for an afternoon of sauna, cold plunges, and acai bowls. Best of all, no one was trying to hard sell to each other
Will definitely be doing these again :)
Interfaze <> Rent a Human signs look awesome!
If you spot one and send me a selfie with the sign, I’ll give you $5000 credits
If you know how low our token cost are, you know how long that would last you…