Tuesday, 29 October 2024

Extending JDE to generative AI

I don't do a lot of work that would assist me create a decent blog these days, especially one for the CNC audience.  Although I think there are a couple of videos in this that will resonate.

I've been able to use and configure an existing Azure template to provide JDE customers with secure access to their documentation and data via Generative AI.  This project (https://github.com/microsoft/PubSec-Info-Assistant) is available to anyone.  It did take me quite a while to have it built and deployed to my own private repo.  But it's there now!


I could probably make this public, but I have uploaded a bunch of data that should not be made private, so in this instance I'll keep it to myself.

The portal gives you the ability to upload any data and then it vectortises the data (Vectorizing data in AI means converting data (text, images, etc.) into numerical arrays or vectors. This transformation enables algorithms to process and analyze the data efficiently, especially in machine learning and deep learning tasks.)  and creates, what is your own personal RAG (In AI, RAG (Retrieval-Augmented Generation) combines information retrieval with generative models. It retrieves relevant data from external sources (e.g., databases or documents) to enhance the accuracy and context of the generated response, creating answers based on both learned knowledge and up-to-date information.) based assistant.

The best way I can understand a technology is to use it in anger.  I uploaded about 250 documents that I had immediate access to, so that I could test the LLM.  It is pretty amazing.  I can ask specific questions about JDE data - and I get some accurate results.  Some is rubbish, but what you learn is that you must get better at prompting.



I think that RISEN is a handy way of remembering how you should prompt.

Anyway, here are a couple of videos that'll show you the portal working over my data and the types of things I was able to ask:

This fist video shows the easy process of uploading data:


The second video is all about asking questions:


I hope this makes sense to you, happy to upload more if needed.

I do want you all to know how much this is costing (I'm going to shut it down this week).



So it's costing 2K a month.  that is very reasonable for a fully operational and productive JDE focussed assistant.  There is some cost fine tuning that can be done too.




No comments:

Extending JDE to generative AI