National Design Studio

Modernizing the interfaces that serve everyday citizens

[National Design Studio @ndstudio]

Seeking hardcore engineers & designers to fix the most broken user experiences on Earth

Apply at https://t.co/8T7aEKO0gM

[FCC @FCC]

https://t.co/mcOXDvFeez

[Rapid Response 47 @RapidResponse47]

https://t.co/3yC9nANCqH

[Rapid Response 47 @RapidResponse47]

https://t.co/3yC9nAN4B9

[Edward Coristine @as400495]

You can now report fraud directly on the https://t.co/e55dXUj1UR platform with a reward up to 30k

[National Design Studio @ndstudio]

Go to https://t.co/hs0dWBlaxg

[National Design Studio @ndstudio]

Now on https://t.co/dok91wPlqW: report fraud against government programs in just a few steps and, if eligible, receive a reward up to $30K

[Jared Zoneraich @imjaredz]

Our two goals with Rampart Edge were
1. Reduce false positives
2. Make it way smaller

We built this model to prevent PII from ever even leaving a user's device on https://t.co/ziIQbmUOlQ. Meaning it needs to be downloaded on every device before chat messages leave

To do this we began with a 50.74MB base model and hill-climbed with the following big wins:
- Reducing from 6 layers to 4 layesr
- Making the word table smaller
- 4-bit weights, shared between layers

The training data was entirely synthetic. We saw big wins by starting with government redacted documents and filling in synthetic PII.

[Edward Coristine @as400495]

It was a great honor to work on Trump Accounts. Consider joining @ndstudio to work on much more just like it

[National Design Studio @ndstudio]

Come ship @ndstudio

[Caelin @caelin_sutch]

🛳️🛳️

This is by far one of the best all-client autocompletes I’ve seen - super fast and incredibly secure. @as400495 absolutely cooked

[National Design Studio @ndstudio]

Now on https://t.co/dok91wPTgu: extremely fast smart suggestions with an all-local dictionary

[National Design Studio @ndstudio]

Now on https://t.co/dok91wPlqW: get a step-by-step plan in a single click

[Edward Coristine @as400495]

Ship ship ship

[National Design Studio @ndstudio]

https://t.co/dok91wPlqW just added dark mode and made chat 5x faster

[Joe Gebbia @jgebbia]

Bigger privacy, smaller footprint. Comes built in to https://t.co/k92ctSDT9f. And available open source for all. Link below. Well done team 👏🏼👏🏼

[Director Michael Kratsios @mkratsios47]

Join us!

[🦀 Josh Seefried @JoshSeefried]

Are you a tech nerd who's tired of government websites that look like they were built on a flip phone?

Come fix it. Apply!

[Jared Zoneraich @imjaredz]

Announcing American Forge.

A six-month residency for hardcore engineers to join the US government.

Created in partnership @ndstudio x @cognition

[Jared Zoneraich @imjaredz]

My first task for the National Design Studio was to make Rampart smaller. It's our frontier on-device privacy model trained to filter all personal user data before it even gets to https://t.co/ziIQbmUgwi

After a lot of hill-climbing and data labeling, I'm excited to launch Rampart Edge along with @as400495. So small the weights fit in a notebook. This will noticeably speed up the load time of https://t.co/ziIQbmUgwi for hundreds of thousands of Americans.

Under budget and ahead of schedule!

[National Design Studio @ndstudio]

Join us to help solve the toughest problems in American government.

Apply to join the American Forge residency.

https://t.co/kjX2mY9ajK

https://t.co/Ctv5PN89ak

[National Design Studio @ndstudio]

Introducing American Forge: a six-month residency for exceptional engineers ready to take on the toughest problems in American government.

Run by National Design Studio in partnership with @cognition

https://t.co/177pZ9bj8W

[National Design Studio @ndstudio]

Four months ago, we built Rampart the on-device privacy model.

Today we are releasing Rampart Edge. It's 10x smaller, more accurate, and runs on almost any device.

https://t.co/uJBjlJytbj

[National Design Studio @ndstudio]

We were able to significantly reduce the download size of the model to just 1.45MB.

Rampart Edge is a four-layer model with one shared set of weights stored in 4-bit. Its word table is a low-rank factorization and under a fifth of the size of the original base model table.

It's open source and public domain: https://t.co/0suBdJ6gVt