Our Services
When a practice has years of drawings and documents on disk, the problem is rarely storage. It’s finding the right file when someone needs it — the revision that matches the site, the detail from a similar job, the PDF that went out with the last tender.
We’ve built indexing for a major architectural firm’s repository: Sphinx-style search and structured pull from AutoCAD where the format gives us something solid to hang an index on. Filename search only gets you so far; getting at layers, titles, and attributes is what makes the archive usable.
We don’t sell OCR miracles or claim every PDF becomes perfect data. We’ve also worked on reading AutoCAD files more directly — there are tools out there; some of that world is still flaky — so we stay honest about what we ship versus what we’re exploring.
Sometimes the person asking doesn’t know the filename, the project code, or the exact phrase that was written down. They know what they mean. Keyword indexes stall there; vector and semantic search start to help.
We’re doing that work in OLLMchat — not as a named commercial case study yet, but as the research stack we can bring to similar document and index problems: ask in plain language, get something relevant from a corpus you already own.