Nomosdoc

I'm an engineer. I came up through aerospace — control integration at Airbus, and building AI systems in the real world at PRAXILIA, where I ran AI-driven ERP workflows handling 1,000+ orders a month.

Two threads kept crossing.

The first was the document problem, and I saw it clearly at Airbus. In big engineering organizations, the real bottleneck often isn't the engineering, it's producing the documents around it. Mission and operation analyses, technical manuals, compliance reports, tender responses. Work that takes real expertise but also weeks of coordination, formatting, and revision rounds. I kept thinking: this is exactly the kind of work AI should be taking off our plate.

The second was that I'd been building with LLMs since GPT-3. Every generation got smarter, but I noticed a ceiling that raw model progress wasn't fixing. LLMs are brilliant at short, sharp answers. They fall apart on entire documents. Ask one for 150 coherent pages and you get fragments: sections that contradict each other, terminology that drifts, cross-references that don't hold. And I became convinced of something that goes against the "one tool does everything" hype: you can't build a general tool that does this well: "you have to specialize it". A system built specifically to produce long, coherent documents will always beat a generalist chatbot at producing long, coherent documents.

So I stopped waiting for a bigger model and built the system instead.

Nomosdoc doesn't ask one model to write 200 pages in one go. It orchestrates specialized AI agents in parallel. The output isn't a draft to fix. It's an editable Word file, submission-ready, in minutes.

It's already being used by teams writing RFP and tender responses, and by teams producing and localizing technical manuals across languages, exactly the kind of work I first saw pile up at Airbus.

Documents that took weeks. Now in minutes. That's the whole point.