AI R&D Engineer

You see what's happening.

Every construction company in North America is being told to adopt AI. Most of them buy a tool, run a demo, and watch it fail in production. The outputs hallucinate. The citations are wrong. Nobody knows why it said what it said. There's no audit trail. The foreman doesn't trust it and goes back to the binder.

The problem isn't the model. The problem is that nobody built the system around it.

That's what we're building.

The project.

Scelta deploys AI inside construction and industrial operations. We interview teams, extract undocumented knowledge, and build systems that replace the binder, the spreadsheet, and the phone call. That work is delivered by engineers embedded in client operations.

The problem we keep running into is the same one everyone building with AI hits eventually. The system works. Until it doesn't. The answer looks right but the source is wrong. The output changes based on how the question is phrased. Nobody can explain why it said what it said. In an office that's an inconvenience. On a job site it's a liability.

We're building the infrastructure layer that sits between a raw AI model and a real operational environment. Something that knows what it knows, knows what it doesn't, and can prove the difference. A system that a project manager can open at 6am and trust. Not because we're asking them to, but because it can show its work.

That system starts as internal tooling. Every AI deployment we build for a client runs through it. It raises the floor on what we can confidently ship. Over time it becomes a product in its own right, one that construction firms across North America can use to deploy AI they can actually stand behind.

You'll be building it from the ground up alongside one other engineer, working directly with the founding team. The architecture is yours to shape. The technical decisions are yours to make and defend. The outcome has to work in the real world, with real clients, under real pressure.

What we're looking for.

Not a title. Not a specific stack.

Someone who has been building with these tools long enough to know where they break. You've shipped something with an LLM at the core. You've hit the wall where it's functional but wrong in ways that matter. You understand that evaluation is not an afterthought, it's the product.

You're comfortable with LLM APIs, OpenAI, Anthropic, others. You understand how retrieval works, why embeddings matter, and what happens when a model confidently returns the wrong answer. You've thought about how to catch that before it reaches a user.

You write clean backend code. Python, Node, or something equivalent. You can design a RESTful API, work with structured schemas, and think carefully about how data moves through a system. You don't need every requirement handed to you. You can read a business problem, ask the right questions, and turn it into something buildable.

You document what you build. Not because someone told you to, but because you understand that undocumented systems don't survive, and that the person reading your notes six months from now might be you.

One to three years of real building with AI at the core. Strong fundamentals in software engineering. The ability to work independently without losing momentum. That's the bar. The rest we can work with.

Bonus if you've worked in environments where output accuracy had real consequences, regulated industries, high-stakes operations, anything where being wrong wasn't just an inconvenience.

The details.

This is a fixed-term contract running April through November 2026.

This is not a dead-end contract. The R&D project is the foundation of a larger platform we're building for the construction industry. The engineer who helps build it will understand it better than anyone. There is a full-time role on the other side of this for the right person, and we're straightforward about that from day one.

Windsor-based is preferred. The construction industry is physical, and some of this