NetApp Revamps Build Infrastructure, Cuts Document Build Times by 90%

Original source: NetApp
This article is an editorial summary and interpretation of that content. The ideas belong to the original authors; the selection and writing are by Streamed.News.

This video from NetApp covered a lot of ground. 6 segments stood out as worth your time. Everything below links directly to the timestamp in the original video.

Learn how NetApp achieved a massive reduction in documentation build times, transforming a multi-day process into a few hours using a modern containerized and automated infrastructure.

NetApp Revamps Build Infrastructure, Cuts Document Build Times by 90%

NetApp has dramatically overhauled its build infrastructure, reducing build times for over a million documents from five days to just one or two hours. The company now leverages a robust stack including Trident for containerization, Kubernetes for orchestration, Ansible for automation, and GitHub Actions for continuous integration, all hosted on Amazon FSX ONTAP. This significant improvement allows for more agile and incremental publishing. This infrastructure modernization is critical for supporting the rapid development and deployment of technical documentation in a complex product environment. By streamlining the build process, NetApp can ensure that its vast repository of customer-facing content remains up-to-date and accessible, directly impacting the responsiveness and quality of information provided to its users.

"If we have to do a full site build, which would normally take five days to build all a million-plus documents, we can either do that in one or two."

▶ Watch this segment — 33:02

NetApp Uses Retrieval Augmented Generation to Enhance AI Documentation Accuracy

NetApp employs Retrieval Augmented Generation (RAG) to significantly improve the accuracy of its AI models, particularly for answering customer queries. This approach involves indexing and chunking millions of documents, then weighting specific aspects to retrieve the most relevant information. By providing the AI with a curated set of documents—acting as an “open book”—the model can identify precise answers rather than relying solely on its pre-trained knowledge. A key challenge for the NetApp team is to train the AI not only on what users can do with their products but also on common limitations or “cannot dos,” such as specific actions requiring prior configuration changes. This nuanced training ensures the AI provides more accurate and practical guidance, preventing users from attempting unsupported operations and ultimately enhancing the utility of AI-powered documentation.

"What we're doing in Retrieval Augmented Generation is essentially saying, 'Here's a set of five documents. We know the answer is somewhere in these documents. Find that answer and only pay attention to the answer in our retrieval.'"

▶ Watch this segment — 15:48

Writers Play Key Role in NetApp's Generative AI Success Through Precise Prompt Engineering

Successfully implementing generative AI requires a combination of coding expertise and highly precise prompting, according to NetApp's Grant Glass. The company has found significant success by integrating technical writers directly into the prompting process, leveraging their ability to articulate complex content standards concisely. Writers contribute by defining specific guidelines, such as the structure of a good lead paragraph, including requirements for imperative verbs and exact sentence counts. This collaborative approach between writing and data science teams is essential for developing robust AI solutions that consistently meet quality standards. By empowering writers to translate their domain expertise into clear, actionable prompts, organizations can ensure that AI-generated content adheres to established stylistic and informational guidelines, thereby bridging the gap between technical AI capabilities and practical content creation.

"The writing team and the data science team working together is really the key for how to make generative AI and its prompting in order to really create a good solution."

▶ Watch this segment — 27:56

NetApp Prioritizes Human Feedback to Enhance AI Model Accuracy and Documentation for LLMs

NetApp rigorously enhances its AI model accuracy by individually reviewing every piece of user feedback. This meticulous process aims to diagnose whether errors stem from the generative AI itself, missing documentation, or inherently confusing concepts within the product. This human-centric approach is vital for testing, continuously adjusting documents, and ensuring content remains relevant and clear for all users. Optimizing documentation for new audiences, such as large language models (LLMs), has become a critical task. Historically, documentation was crafted for human readers or search engine optimization, but now it must also cater to how AI systems interpret and synthesize information. This requires a holistic review, sometimes involving changing titles or restructuring content, to ensure that existing materials, like the extensive ONTAP documentation, are effectively consumed by these new AI audiences.

"It's not an automated process. It's figuring out where are the points of failures, but it's also a holistic effort to say, well, we're also going to change the way that the AI thinks about the underlying documentation."

▶ Watch this segment — 19:08

NetApp Plans AI-Powered Best Practices and In-Product Assistance Through Data Integration

NetApp is exploring future applications for its AI by integrating its extensive documentation dataset with customer auto-support features. The goal is to provide proactive best practice recommendations based on real-world customer usage patterns. By combining these data sources, the AI can analyze how users interact with products and suggest optimal configurations or actions before issues arise. These AI-driven recommendations are intended for direct integration into NetApp products, such as BlueXP, offering in-product assistance. Collaborating closely with UX designers, the company plans to syndicate this intelligent guidance into user interfaces, enabling customers to receive timely advice and troubleshoot potential problems without leaving their workflow. This initiative aims to transform reactive support into proactive, intelligent assistance.

"We're imagining some kind of experience where our content is essentially syndicated into and combined with the product itself to guide you down the right tree."

▶ Watch this segment — 38:59

NetApp Navigates AI Training Challenges with Over 120,000 Complex Documents

Training AI on NetApp's extensive documentation, which includes over 120,000 documents outside of knowledge bases, presents significant challenges due to products often sounding similar. This requires extensive work to disambiguate the unique value propositions of various offerings and to navigate diverse user language and intentions to provide relevant product information. The sheer volume and complexity of the content demand careful model tuning. Generative AI, however, provides a direct line into understanding customer questions and pain points, offering invaluable insights for documentation improvement. By observing how users interact with the AI and what queries they pose, NetApp can identify gaps or areas requiring more detail in its documentation. This feedback loop allows for continuous refinement of content to better serve technical users and address their specific needs.

"What are our users asking for, what are they interacting with our products, what are their pain points? And then we can use that to go back to the documentation and say, 'Where can we add more detail to help our customers?'"

▶ Watch this segment — 12:47

Also mentioned in this video

• NetApp's documentation has been revamped, making it faster and more accurate,… (0:00)
• NetApp docs transitioned from a closed, XML-based system to an open,… (3:37)
• Jen Kaufman clarifies that the core purpose of technical documentation remains… (6:23)
• The positive synergy between NetApp docs and the knowledge base (KB) team,… (9:26)
• NetApp's AI assistant, Doc (named after their owl mascot), is trained on… (11:26)
• Fine-tuning involves adapting to customer language, such as recognizing… (18:15)
• His team's goal is to build an AI landscape supporting various AI processes… (23:23)
• They initially avoided embeddings due to cost and frequency of republishing,… (25:03)
• The authoring assistant, a plugin for GitHub Copilot, which his team developed… (25:54)
• Axel Davis adds that writers play a crucial role in evaluating and training… (30:08)
• Leveraging NetApp technologies internally, noting experiments with both… (30:56)
• NetApp navigates IP and legal challenges by leveraging the fact that their… (33:47)
• Their partnership with Microsoft and two years of working with Azure OpenAI… (36:06)
• A shift in mindset for the NetApp docs team from content creators to data… (37:15)
• Adam Newton mentions that the globalization team is using AI to evaluate and… (41:17)
• Plans to incorporate search and AI functionality directly on NetApp docs pages,… (42:29)
• Generating content from engineering specs to create AI-driven synopses and… (43:12)
• The organization's focus on intelligently applying AI to improve business speed… (44:17)
• His team's biggest goal is to improve the findability of NetApp docs,… (45:46)

Summarised from NetApp · 48:43. All credit belongs to the original creators. Streamed.News summarises publicly available video content.