Why Ernest rather than ChatGPT, Gemini or Claude to read an inspection report

You can. Many people do, it is like asking your brother-in-law, and for a one-off question it is more than enough. Deciding on a $500,000 purchase from 80 pages is something else.

I am not a chat with a good prompt: behind the conversation there is an analysis engine, built on more than a year of work with building inspectors, on their report-writing software. Here is what changes.

Behind the conversation, there is an engine. A general assistant receives your PDF, reads part of it, summarizes the rest and answers with great confidence. I chain specialized steps, each with its own rules: identifying the report and its severity legend, reading a few pages at a time in parallel, mechanically verifying quotes, classifying, consolidating, then writing. Reproducing that in a chat means dozens of prompts, manual checking page by page, and expertise you would already need to have.

I read everything, and prove it without AI. Every inspector finding is captured with its page and its exact sentence. A deterministic verifier, with no AI model at all, then checks every quote against the text of the stated page. What cannot be found in the report is dropped before it ever enters your analysis. A chat cannot make that promise: it has no mechanism to keep it.

I speak the inspectors’ language. Every finding is classified on PRISM, Lumos’s building taxonomy (160 categories across 28 domains, from foundation to ventilation), then qualified under the BNQ 3009-500 grid: observed apparent defect, sign of deficiency to confirm, safety risk, limitation, information. A chat pours all of that into one list of “issues”. I keep those distinctions and carry the severity label set by your inspector, never contradicting it.

I know your market. Estimates are produced with cost grids per region (Québec, Ontario, British Columbia, Alberta, rest of Canada, United States), in your currency, with the vocabulary of the trades here: master electrician, master plumber, licensed roofer, the inspection condition of the purchase agreement, the syndicate’s records for a condo. When the report first calls for an expert, I do not price it and I tell you.

I do not answer a question, I prepare you for a decision. A chat gives you an answer, then you prompt twenty more times to get the rest. I deliver the whole analysis at once: a justified rating out of ten, three lists (“Fix first”, “To check”, “Can wait”), the repair estimate, the negotiation strategy, the first-year roadmap and the questions to ask your inspector.

I am measured, then guaranteed. Lumos measures the engine on a corpus of real reports: recall of deficiencies, quotes found, stability from one run to the next. And if you show, with evidence, that I got it wrong or that the analysis is not relevant to your report, Lumos refunds 100%. No general assistant measures or guarantees what it tells you.

Here are the options side by side.

To understand a term, rephrase a paragraph or check a hunch, ChatGPT, Gemini or Claude do the job well. I am for something else: reading a whole report without letting anything through, running it through the inspectors’ grid, and handing you a plan to decide.

I do not visit the house and I do not replace your inspector.

I do not produce quotes: amounts remain orders of magnitude.

I give no legal opinion on latent defects or warranty.

I do not guess what is not in the report. Limitations remain unknowns, but you will know how to close them.

When you are ready, attach your PDF report.