Ai Engineer

Vedi l’offerta su Go Lavoro: golavoro.it/annuncio-lavoro/7atrv0

Robusto builds AI tools that automate professional workflows on macOS. Our software runs on device, with no cloud dependency, and is used every day by engineers, studios and post production facilities. We're based in Turin at I3P, the Politecnico di Torino incubator, and we work hybrid. We hire fast: two interviews, maximum one month from application to offer.
Our tech stack
Python, Swift, C++ where it earns it. Computer vision and machine learning, model training and on device inference, function calling protocols, macOS accessibility and input APIs. We'll walk you through the rest in person.
About the role
We're building the layer that lets an AI operate professional desktop software that exposes nothing readable to a machine, on computers that cannot or will not send data outside.
Detection is not the hard part anymore. Meaning is. Knowing that a group of pixels is a button is easy. Knowing that it's the export button, that it belongs to that panel, that it's currently disabled, and then handing all of that to a model in a form it can act on without burning half its context, that's the work.
This is not a research role and it's not a wrapper around someone else's API. You'll own a piece of the engine end to end and watch it run on real machines doing real work. You'll work directly with our CTO and, as the team grows, become the reference point for the engineers who come after you.
What you'll work on
Screen understanding.
Grouping elements into panels and lists, tying icons to their captions, inferring what a control does from how it looks, reading state from pixels. Classical computer vision where it wins,



edges and connected components and adaptive binarization and template matching. Getting text out of dense, dark, low contrast professional interfaces that were never designed to be read by a machine.
Models we train ourselves.
Classical methods run out, and when they do we train. Building the datasets, deciding what a learned model buys over the algorithm it replaces, and getting it small and fast enough to run on the user's machine.
Tool design for models.
Naming, descriptions, error messages that tell the model what to do next instead of just what went wrong. Half our wins last month came from rewording a single tool result.
Agent loops that recover.
Multi step execution with budgets, loop breaking, verification of the effect after each step, and honest failure instead of confident nonsense.
Grounding, memory and evaluation.
Retrieval over accumulated knowledge of an application, so the model's context stays small and textual. And benchmarks across deliberately different applications, so an improvement generalizes instead of overfitting to whichever app you happened to be staring at.
What we're looking for
You've built agents and watched them fail.




And you came out of it believing that most failures were bad tool contracts and missing verbs, not a weak model. You can't learn that from a paper.
Real computer vision fundamentals, not a model zoo.
You'll reach for thresholding and connected components when that's the right answer, and you can defend the latency of a neural network when it isn't.
You're pragmatic about the platform.
Whatever the OS hands you is a tool, not an identity.
You measure.
You build the evaluation before you claim the improvement.
Python fluency.
Swift or C++ is a plus, not a filter.
Nice to have
MCP or other function calling protocols in production. On device inference and model optimization. Dataset and annotation pipeline work. Anything you've open sourced.
Two seats, two levels
We're hiring two people into this role. One seat for someone earlier in their career with the fundamentals and the obsession, one for someone senior who has already shipped a system like this and can set direction without being asked.
€28k to €50k RAL depending on level and experience, plus equity. We'll tell you which seat we think you fit after the first call.
Where and when
Turin, hybrid. We're open to people based elsewhere in Europe who can be in Turin regularly. This is not a fully remote role.
Start as soon as you can. We're moving on this now.
How our hiring works
Every application is read by a founder.
Week 1
First call with one of the founders (30 min)
Week 2
Technical interview with the CTO (60 min)
Week 3
Offer
Maximum one month from application to offer.

Vedi l’offerta su Go Lavoro: golavoro.it/annuncio-lavoro/7atrv0