SafetyKit replaces human content moderators with language models. Companies like Character.ai, Substack, Upwork, Faire, and Eventbrite use SafetyKit to find and remove bad stuff on their platform without hurting good users.

We’re ex-Stripe and Airbnb, repeat founders, and have deep connections and experience in our space. We work together in-person every day in a sunny office in the Mission in San Francisco.

Find your maximum velocity.

SafetyKit is group of engineers, designers, and operators committed to deploying AI into the world's largest companies as quickly as possible.

It's going to happen, and we're not going to watch it from the sidelines.

We're obsessive builders. We couldn't stop coding if we tried. We chose SafetyKit over cushy big tech jobs because we're tired of moving at the speed of consensus. We want to know what our maximum velocity is.

We're not afraid of big customers, high scale, and high expectations. We know that impact is denominated in dollars — we care about creating real, visible economic change inside the world's largest companies.

We want to learn from the best. We're former and future founders. Some of us are pedigreed (Stripe, Netflix, MIT, Yale), some of us aren't. Some of us are 14-year staff engineers, some of us are on day one.

It's working — some of the biggest companies in the world depend on the work we do every day. But it's still early. 99.9% of the value we'll ever produce is ahead of us.

If this sounds like the life you want to live, let's talk.

What we care about

Speed up.

The future is waiting on us. Don’t do tomorrow what you can do today. Find the minimum effective dose and move on to the next problem.

Obsess.

Reality has a surprising amount of detail. Go deeper. Love your work — even the stuff that other people find unbearably boring

Decouple and trust.

We are a tightly-bound team of autonomous operators. We maintain a high bar so we can move fast with high trust. We support and teach each other every day.

Some of our past interns

• Built AI agents to autonomously investigate signup flows
• Built a classifier to determine if marketplace products are drop-shipped
• Built a classifier to identify IP violations deployed on some of the world’s largest marketplaces

About you

• You love language models.
• You code and ship at very high velocity.
• You aren’t scared to use codegen models to code and ship even faster.
• You act like an owner in everything you do.
• People tell you you pick things up unusually quickly.
• You care about your work and you care about moving fast. A lot.
• You know more than you need to know. You have an insatiable curiosity.

We build on AWS serverless (Lambda, Step Functions, SQS), usingTypescript. It's fast and fun.

We use open source language models, big hosted models, specialized vision models to make users happy. We merge, eval, and ship new models the day they launch (if evals look good!).

Every engineer at SafetyKit works to understand our users' businesses and policies inside and out. That lets us push available tech to its limit and ship exactly what our users need faster than anyone else in the world.

We love managing complexity so our users lives can be simple. We have in-house eval tools, tools for debugging complex LLM execution chains, and lots of LLM-assisted code and infra generation.