Roschiwal + Partner: Making a Large Document Archive Accessible, Drawings Included
Data that has grown over years is not a search problem, it is an access problem. How Roschiwal + Partner put its own archive to use, and why CAD and drawings are the harder part.
Roschiwal + Partner is an engineering firm with a very large internal document archive that has grown over years. KoAssist makes that archive searchable and backs every answer with a file and a page. Together, first steps have been taken to bring CAD data and drawings into scope as well, not just running text.
An archive that has grown beyond anyone's overview
Roschiwal + Partner has worked as an engineering firm on a large number of projects for years. What comes out of that stays: calculations, specifications, reports, correspondence, drawings, project files. The archive has grown very large over time.
Large volumes of data are not a problem by themselves. They become one the moment nobody has a full overview of what actually exists anymore. At that point, even full-text search stops helping, because it assumes you already know the right term. That is exactly what you do not know when you have a question about an unfamiliar or old project.
From searching to asking
KoAssist draws on the existing archive and answers questions from it, instead of returning a list of hits. The difference is practical: you phrase the question the way you would ask a colleague, and get an answer with a reference to the file and the page.
That reference is not decoration. In an engineering firm, a statement without a source is not usable, because it cannot be checked. Only the evidence turns a plausible-sounding answer into one you can actually work with.
Why drawings are the harder part
In technical documents, a significant part of the information is not in running text. It sits in dimensions, in sectional views, in symbols, in title blocks, in the spatial layout of the sheet. A system that only pulls text out of a file sees almost nothing of a drawing.
That is why drawing integration is not a side issue. It is the point that decides whether document search really holds up in engineering.
First steps into CAD and drawings
First steps have been taken together with Roschiwal + Partner here. This is ongoing work, not a finished feature, and it is deliberately described that way: drawings are a hard field, and presenting a work in progress as finished would not help anyone.
The foundation is in place, because KoAssist works multimodally and evaluates image content instead of skipping over it. What becomes of that foundation in practice takes shape through the collaboration.
Related topics
Why images matter as much as text in engineering is covered in the article on multimodal AI in engineering. For an overview of the use cases, see the page AI in Engineering Design; for details on hosting and data protection, see the page Security.
FAQ
From what volume of data does an AI-powered search pay off?
Volume alone is not the trigger. What matters is whether the knowledge can still be traced through who is responsible for it. Once nobody has a full overview of what exists anymore, full-text search stops helping much, because it assumes you already know the right term to search for.
Why are drawings harder than text documents?
Because the information is not in running text, it is in dimensions, symbols, title blocks and the spatial layout of the sheet. A system that only extracts text sees almost nothing of a drawing. It takes a model that understands the image itself.
Does the data have to leave the company?
No. KoAssist is hosted in the EU and operates in compliance with the GDPR. Customer data is not used for training. Details are on the Security and data protection page.
What is the current state of the CAD and drawing integration?
First steps have been taken together. This is ongoing work, not a finished feature. The foundation is in place, because KoAssist works multimodally and evaluates images just as it does text.
A chatbot for technical documentation: what it actually has to do
A chatbot on your own technical documentation sounds straightforward. In practice its usefulness is decided at four points, and generic chatbots fail at all of them.
What is design automation?
The term covers very different things, from a parametric template to a full configurator. A look at the stages, and at the precondition that is almost always underestimated.
ChatGPT for internal documents: secure and GDPR-compliant?
Many teams want to point ChatGPT at their own files. What risks this creates, when the use is GDPR-compliant, and how a dedicated solution differs from the public chatbot.
Less searching.
More engineering.
In a 30-minute demo, we show KoAssist working with your own documents and discuss setup, integrations and pricing for your team size.