Knowledge you already own
The answers are already in your files. The problem is that finding them depends on one person remembering which folder, which year, which version. When they are busy or gone, the archive goes quiet.
The goal is to make the knowledge askable: a plain-language question in, a sourced answer out, from files you still own.
§1 Why search alone does not solve it
Search returns a list of documents that contain your words. It cannot tell you which version was approved, or read across three files to give you one answer. Someone still has to open each result and piece the answer together, which is exactly the work you were trying to remove.
A general chatbot pointed at your files will answer even when the answer is not there. It fills the gap with something plausible, and you cannot tell the difference until it costs you. The system has to be able to say it does not know, and it has to show its source every time.
§2 How I build it
You ask in plain language and you get an answer that names its source file and page. You open it and confirm. This holds for every answer, including the ones that are wrong, so a mistake is always traceable rather than hidden. You can check where every answer came from, and that is what makes the answers worth trusting.
Before it is done, you give me ten real questions. I do not write them, you do. Eight of ten must be answered correctly, and you decide what counts as correct. Every answer cites its source, and where there is no answer the system must say so. That is the bar the build has to clear before you rely on it.
§3 What it can and cannot do
This matters more than the price, so I say it before you start, not after. It answers well within one topic and one period. It answers across several years with more care, and very large archives get scoped in stages. It reads the text inside your files, so it can name an approved design and point you to the page, but it cannot display the artwork itself. Where there is a real limit, you hear it from me first.
§4 To start
Tell me roughly how many documents you hold, where they live, and the kind of question you wish you could ask them. We prove it on a small slice first, on real questions you write, before scaling to the full archive.