The Reasoning Show

The Reasoning Show AI moves fast. Thinking clearly matters more.

The Reasoning Show cuts through the hype to explore how the smartest people in enterprise AI actually make decisions — the strategy, the tradeoffs, and the hard lessons no press release mentions.

Every week, hosts Aaron Delp and Brian Gracely sit down with the founders building the tools, investors funding the shift, and operators running AI in the real world. Not hype. Not panic. Just clear-headed conversations with people who have to make actual decisions.

Because the AI revolution isn't just happening. It's being reasoned through.

New shows every Wednesday and Sunday.

Topics: Enterprise AI strategy · LLMs in production · AI leadership · Agentic AI · Digital Sovereignty · Machine Learning · AI startups · Cloud Computing

All Episodes

The Reasoning Show

RAG That Survives Production

January 14, 2026 • Massive Studios

0:00 | 22:22

Adam Kamor (@tonicfakedata, Founder & Head of Engineering) talks about building RAG (Retrieval Augmented Generation) systems in production for AI data.

SHOW: 992

SHOW TRANSCRIPT: The Cloudcast #992 Transcript

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

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

SHOW NOTES:

Topic 1 - Adam, welcome to the show. Give everyone a brief introduction.

Topic 2: Our topic today is RAG systems, specifically RAG in production. Let’s start with customization sources and types. When it comes to customizing off-the-shelf LLMs, RAG is one option, as is an MCP connection to a SQL database, and there is pre- and post-training, as well as fine-tuning. How does an organization decide what path is best for customization?

Topic 3 - RAG came on the scene as the savior for organizations that want to use customer AI without the need for fine-tuning and additional training. It has either gone through or is currently still in the trough of disillusionment. What are your thoughts on RAG's evolution and the challenges it faces?

Topic 4 - Let’s walk through the basics of validation. Once you set up RAG, how would an organization know it works? How is accuracy measured and validated? Are you looking for hallucinations? Context quality?

Topic 5 - What is Tonic Validate, and where does it fit into this stack? Is it in band? Out of band? Built into the CI workflow?

Topic 6 - Accuracy is one aspect, but we hear more and more about ROI for Enterprises. How should ROI, risk, and compliance be measured?

Topic 7 - Where and how does security fit into all of this? Also, your thoughts on synthetic data for training vs. real data?

FEEDBACK?