Founding Engineer - Full-Stack & Infrastructure

Founding Engineer - Full-Stack & Infrastructure · Haifa · Hybrid

We built a system that predicts how a lawsuit will end — settlement value, escalation risk, trajectory - and insurers make real reserving decisions on its output. It works. It got here fast, and what comes next is a bigger engineering problem than the one that got us here. That's where you come in.

That's the job.

WHY "FOUNDING"

We're small. There's no engineering org to join, no process to inherit, no PM between you and the problem. The conventions this company runs on for the next five years -  how we test, how we deploy, what we build on - are mostly not decided yet. You'd be making those calls with us

You'll work directly with the founders. Not a reporting line through three people.

WHAT YOU'D ACTUALLY BE DOING

The stack: Python/FastAPI microservices on GCP, PostgreSQL, Terraform, React 19 + TypeScript, an OCR pipeline running GPU workers over scanned legal documents, and AI pipelines that turn 400-page case files into structured, explainable predictions.

Real problems on the table right now, not hypotheticals:

· How do you test a system whose core component is non-deterministic? We have opinions. They're probably wrong. We want yours.

· Our document pipeline reads legal PDFs that were scanned in 1998. Accuracy is a model problem, a preprocessing problem and a cost problem at once.

· Every prediction must be explainable to a claims adjuster who may be cross-examined about it. That constraint reaches all the way down into architecture.

· We're at the stage where the right abstraction saves a year and the wrong one costs one.

You'd own systems end to end - schema, service, pipeline, deploy — including the part where you're the one who gets paged when they break.

WHAT WE'RE SCREENING FOR

Not years. Three years in and you've shipped things that scared you a little? Apply.

What we can't teach:

· You ship. Things you built are running in the world, and you can walk us through why they're built that way.

· You go and find out. When something breaks in a layer you've never touched, you read the source instead of waiting for someone who knows it.

· You say what's true. Including "I broke it," "I don't know this," and "I think this plan is wrong, here's why."

· You finish. The unglamorous 20% - the migration, the error path, the runbook - gets done, because that's where systems actually fail.

· You can live with what you build. Shipping it is half the job; running it for two years is the other half, and you want both.

Strong Python is the one hard requirement. Everything else on our stack you can pick up here if the above is true. We'd rather hire someone who'll learn Terraform in a month than someone who lists it and doesn't care.

HONESTLY, THOUGH

Small team. No PM will hand you a spec - you'll often be the one deciding what "done" means. Some of what you inherit you'll want to rewrite; some of it you won't be allowed to yet, because it's serving customers. If you need a defined lane, this will be uncomfortable.

Hybrid: 2 days a week in our Haifa office.