The roofer building an AI that refuses to take sides

The moment that started it happened in a homeowner's living room. A carrier adjuster had just told Andrew Stevens's client she owed two deductibles, because, he claimed, two separate storms had somehow struck different halves of the same house. Stevens asked the obvious question, right there in front of her: which storm came first, and what was the exact date of the one he was sure about.

The adjuster froze. His answer, Stevens recalls, was “this is why we work in tandem.” It was, in effect, an admission that he had simply wanted a second adjuster in the room to outnumber the contractor. Stevens fought the denial that followed using straight Newtonian physics, and an engineer came out and verified his math. The carrier stonewalled anyway. They got away with it because the language Stevens had used crossed a line, into what the industry calls the Unauthorized Practice of Public Adjusting, or of Law.

I'm tired of seeing homeowners penalized by technicalities.

A game played with language

The property claims process, as Stevens describes it, has quietly devolved into gamesmanship. Carriers stall, waiting for a contractor to slip and utter a phrase that violates the rules on who is allowed to argue a claim. Contractors, for their part, lean on emotion and argument instead of sticking to regulated codes and manufacturer mandates. Both sides end up playing a semantic game, and the person who loses it is almost always the homeowner, cut out of coverage they had paid for.

Coming from a background in HVAC and roofing, Stevens fought this on the front lines for years. Loretta Compliance is his attempt to change the terms of the fight: a forensic auditing platform built specifically for insurance and property claims compliance, with a live codebase built on Python and FastAPI.

Why the easy route was off the table

The obvious way to build it would have been a wrapper around a cloud AI model. Stevens scrapped that early, once he realised what a single hallucination could cost. In this industry, one invented fact could expose a contractor to real legal liability, or cost a homeowner their roof. So he engineered something harder: localized machine-learning execution environments running on tools like Ollama and LM Studio, locked down with custom anti-loop and anti-drift guardrails that force outputs based purely on empirical building codes.

Neutrality holds no opinions.

That phrase has become the platform's core philosophy. People carry biases; standard AI models carry them too. Loretta Compliance is built to remove them, to audit a claim against codes and manufacturer mandates rather than against whoever is more willing to bluff.

The fight he did not expect

The hardest part, Stevens is candid, has not been the architecture. It has been believing he is allowed to build it. Going from roofing to designing a machine-learning compliance system, he says, messes with your head. There are days he looks at his own system and convinces himself he has built nothing of value, that he just talked an AI into confident nonsense. When that hits, he goes back to the PDF outputs his engine generates and checks the work. The math holds. The guardrails hold.

His belief, hard won, is that you do not need a computer-science degree to solve a technical problem of this kind. You need to understand the physical reality of it better than anyone else. The tech stack can be learned and engineered. The years spent arguing claims in someone's living room cannot be faked. If Loretta Compliance works the way he intends, that neutrality becomes the standard ground the whole industry has to stand on.

This profile is drawn from Andrew Stevens's own account of building Loretta Compliance, an early-stage venture in active development. Product details are as described by the founder.