Using off-the-shelf vs. fine-tuned models

Stephan Donze (@sdonze CEO @AODocs), discusses the enterprise unstructured data crisis, where 80% of business data remains untapped due to legacy system limitations and the challenges of AI-powered document management at scale. We explore how AI agents can transform document workflows while maintaining trust and compliance, the architectural principles needed for cloud-native document management, and why traditional search fails in the age of generative AI.

SHOW: 961

SHOW TRANSCRIPT: The Cloudcast #961 Transcript

SHOW VIDEO: https://youtube.com/@TheCloudcastNET 

NEW TO CLOUD? CHECK OUT OUR OTHER PODCAST - "CLOUDCAST BASICS"

SPONSORS:

• [Interconnected] Interconnected is a new series from Equinix diving into the infrastructure that keeps our digital world running. With expert guests and real-world insights, we explore the systems driving AI, automation, quantum, and more. Just search “Interconnected by Equinix”.
• [TestKube] TestKube is Kubernetes-native testing platform, orchestrating all your test tools, environments, and pipelines into scalable workflows empowering Continuous Testing. Check it out at TestKube.io/cloudcast

SHOW NOTES:

• AODocs website

Topic 1 - Welcome to the show, Stephan. Give everyone a quick introduction.

Topic 2 - We hear all the time about unstructured data and the continual growth in the Enterprise. I’ve heard numbers of upwards of 80% of all corporate data is unstructured. I’ve worked at several companies and supported a significant number of customers over the years, and I can count on one hand how many say they have “control” of their data. How did this come to be, and is the problem as big as I think?

Topic 3 - The second part of this, and this might be an even bigger problem, is how much of the data is used? Too many needles in the haystack, if you will. How does Agentic AI address this challenge, and where do traditional document management systems fail?

Topic 4 - We’ve talked about data quality in the past on the show, and I’m wondering if this also becomes an issue. Let’s say you have a bunch of draft documents leading up to the final version. Is it possible that improper version control and/or we’re back to a data quality problem of finding the “final version” needle in the haystack? How does AI prevent this and also not hallucinate an answer that may not be true?

Topic 5 - Some have called AI’s ability to absorb and report on data just fancy search. What are your thoughts on this? Where and how does traditional search differ from Agentic AI management?

Topic 6 - I also see this as being so much more than indexing and reporting on documents. There is also the concept of automation and workflows that agentic AI can improve upon. What use cases are your customers implementing?

Topic 7 - Where do you think the industry will go in the next 2-3 years?

FEEDBACK?

• Email: show at the cloudcast dot net
• Bluesky: @cloudcastpod.bsky.social
• Twitter/X: @cloudcastpod
• Instagram: @cloudcastpod
• TikTok: @cloudcastpod