I want to suggest using a tool like privateGPT, trained using all these rules to at least make searches more "humane" than knowing the exact product code or making someone face a wall of text when you just want to know a specific rate or sale restriction.
However, since it's a state owned company, things move REALLY SLOW. Slow as in "the modern chatbot they tested last year just repeats 'how can i help you' until you say one of the five initial terms without typos and god knows how many thousand dollars they burned on this, so sunk cost fallacy keeps it alive" slow. There is a good chance that if i just suggest it, the IT will say it has a project related to it, but it'll be stuck for a decade in management hell.
So i considered borrowing the wiki and making a working chatbot (with privateGPT[1] on an offline notebook. Slow, but very low data leak risk), for example, and presenting it as a proof of concept directly. On one hand, it would drive the point across or i would learn first hand it just doesn't work. On the other, it is a policy violation punishable with instant firing.
And no, sales department grunts don't get their own machines or access to anything of the sort. And if i tried to reach someone high enough on hierarchy that they can formally request the hardware and data to make this, it's their project now and it has to compete for resources with every other project from people for the same department and level, which causes the slow progress i told before.
So... How do i advance? Go ballistic and risk having my head cut off (it's less appealing once you have a family, bills to pay and a mortgage), give it away, renounce any credit and pray it moves faster than the average innovation or is there another way of making a prototype and prove it can help.
I considered loading another wiki of something different, but it's less convincing as a proof of concept because it needs to allow addition, change and exclusion of small bits of data without the bot going offline for retraining 3 times a day. I can sideline it if i show it answering work related questions and solve it later, but with an external example, they'll possibly brush it off as not even showing "useful info".
Tnaks for your input
[1] https://github.com/imartinez/privateGPT
I'm not sure how privateGPT works, but any sane implementation is going to store the data in a vector database, not constantly train it into the llm itself. Updating it is easy, fast, and can probably be done without going offline.
...But yeah, the key is somehow replicating your wiki closely enough for a demo. With access to the wiki, maybe you could copy the structure (but not any data) and populate it with dummy data from an LLM? If they are good at anything, its producing something convincing and totally made up.
EDIT: Also, if the backend is llama.cpp, recent advances will make it very easy to run, but questionable for commercial use depending on the model it uses. LLaMA itself is basically illegal to use, even if Facebook doesn't seem to care.