We’re Jetty.

We’re a small team building the tools that let you delegate real work to AI and know whether it was done well.

Our approach

As ML engineers hired to run evals, we never had a good tool that works for everyone. So we’re building one.

01
Evaluation is a key part of AI automation
Handing work to an AI agent will ultimately require a human to check it. However, today most tools make that work a side quest. Instead of running evals outside, Jetty runs the work and checks it at the same time, so a bad run gets caught and we can make the next one better. The best part is that hard questions like “what is this costing me?” or “is this the right model for the job?” become a lot easier to answer.
The workloads we do often change when you least expect it, and so tracking inputs and outputs mean that mistakes become calibration examples for the future.
02
The expert stays in control
The person who knows what a good result looks like is the person who does the job. We think that to get really good results from AI, we need to build tools so experts are in the loop.
03
Workloads should be easy to write and share
The job to be done should be written down in plain language so you can read, diff, and version it as you see fit.

We’ve felt the pain of AI Whack-a-Mole

We built Jetty around the run: everything an AI workload produces is kept in one place, so when a result is off you can find out why. We also run Jetty on Jetty and are constantly working out the rough edges for ourselves and our users.

Bagels, blizzards and Mila

We’re based in sunny and sometimes snowy Montreal, Canada. We work out of the Mila Ventures space, a hub for AI startups inside the Mila research institute. We chose this place because it’s the happiest place we know with researchers, whiteboards, and the sense that the next breakthrough is happening just down the hall.