Rob Lambert

I'm a self-taught builder in Ottawa making practical AI tooling for the document-heavy, regulated work that government contractors still do by hand. What I'm building now is a deterministic engine that reads an RFP or Statement of Work and pulls the binding "shall" and "must" language into a page-cited matrix, while separating out the evaluation criteria, definitions, and submission rules that look like requirements but aren't. The goal is that every row traces straight back to a quote on a specific page in the source. I optimize for "does it work on a real document" over "does it demo well." I haven't yet benchmarked it against a hand-built matrix — that's the validation I'd want before anyone trusts it on a live bid, and I'd rather say that plainly than oversell it. I also ship and maintain an MIT-licensed Python package on PyPI (pip install nano-empire-tollbooth) — so the work is installable, not vaporware. I'm not selling anything. I'm trying to learn what would make a tool like this genuinely trustworthy to a proposal team. If you've felt the pain of the first-pass read on an RFP, I'd value 10 minutes comparing notes.