IBM watsonx Assistant for Z is generative AI solution, transforming and simplifying the way Z users of all experience levels engage and interact with the mainframe to be more productive. Users can ask anything, any time, as many times as you need and your Z assistant is there to provide accurate and current information that leverages all IBM Z domain-specific knowledge. Furthermore, it uses granite-13b-chat-v2 LLM that meets stringent governance and compliance criteria
For those of you who have not had a chance to learn about the product, I encourage you to check out this case study from our own IBM CIO that is published here . We refer to our IBM CIO office as Client 0 since, with over 200+ internal applications that run on 600+ LPARs across 4 data centers, they share the same challenges and opportunities that many of our enterprise customer experience. Our CIO Office is the first to try out many of our new products and offer insights on how you can benefit from many of these solutions.
One of their top requirements for watsonx Assistant for Z, the most requested capabilities from our customers and sponsor users, was the ability to Bring Your Own Documentation (BYOD). Clients viewed the ability to bring their enterprise mainframe process, application, and operations documents in addition to the IBM documentation as essential to customize the product for an enterprise roll-out.
Why is this one requirement so important?
Mainframes have existed for 60 years and are continuing to thrive and be the most important business asset for many of our customers.
With such a rich history, over the years, enterprises have built their critical applications on it and processes around it. Most of this knowledge is maintained in documents, which have been built, fine-tuned, and passed down for decades. With the evolution of document management solutions over the years, many of these golden documents have moved homes multiple times and stayed accessible, while some still reside in remote servers or worse, user workstations.
The tribal knowledge contained in these docs are essential to provide understanding of system topology, application and process descriptions adopted by the enterprise, which can benefit new mainframes by providing history and context.
While IBM Documents, Redbooks, Redpapers, support docs, etc. provide the current state of products, solutions, and best practices, it is your enterprise specific documents that explain the structure and adoption of these tools, best practices, and solutions by an enterprise.
Hence, bringing these enterprise-personalized documents into the corpus of the Z RAG (1) to be mined in addition to IBM documents when providing curated answers to the incoming questions make the response immediately usable and relatable to the enterprise's unique adoption and usage of the mainframe. With that context, I am excited to share that all that is now possible in V2 and beyond.
A key feature of this release is the capability to allow customers to bring their enterprise documentation into the corpus of documents referenced by the Z RAG(1) component shown below, enabling clients to personalize their Z RAG(1) by ingesting proprietary and entitled 3rd party documentation. Now, when users ask a question, IBM watsonx Assistant for Z leverages the Z RAG to ground responses with both IBM Z and your enterprise knowledge.
Figure 1: IBM watsonx Assistant for Z – V2.0 Architecture
The process of document ingestion is a simple 3-step process done through a command line utility, zassist.
- Point zassist to the directory with the documents on prem
- Initiate the document ingestion process
- Refresh the ZRAG to update references to the new documents ingested
The following file formats are supported - docx, html, and pdf. As you would expect, the time to ingest is affected by the size and quantity of documents. However, running this process on GPU capacity is much faster than running on CPU capacity. So, if you have that spare GPU capacity you can put it to skillful use to run the document ingestion like a nightly batch job.
Currently, we have benchmarked optimal performance for 15,000 Files with 50 MB per file at a max of 100GB. This number will however continue to evolve to suit the enterprise client's expectations as we continue to work closely with our early adopters and sponsor users.