Teaching AWS in the Real World at AI Con USA 2026

By Scott Peterson and Tim Collinson, senior software consultants

This past week we were in Seattle for AI Con USA 2026, co-teaching a half-day, hands-on tutorial: "AI Deep Dive: Exploring AWS Using Real-World Scenarios." Three and a half hours, a room full of engineers, data scientists, analysts, and technology leaders, and one goal: get people building real AI on AWS instead of just hearing about it.

Here is what we taught, and a few things we hope the room walked away with.

Building on Amazon Bedrock

We started with Amazon Bedrock, AWS's managed service for foundation models. The point we kept coming back to: Bedrock is the easy part. Calling a model is a few lines of code. The work that actually matters is everything around it, getting your data to the model, grounding its answers, and handling the cases where reality does not match the demo.

Grounding answers with Knowledge Bases

From there we moved to Bedrock Knowledge Bases, which is where a generic model becomes useful for a specific business. Instead of hoping a model already "knows" your domain, you point it at your own documents and data and let it retrieve the relevant pieces at query time. The room got hands-on connecting datasets, building a knowledge base, and watching the difference between an ungrounded model guessing and a grounded one answering from your actual data.

Working with real datasets and financial data

The scenario we built around was analyzing financial data, which is a deliberately unforgiving test case. Financial data is messy, high-stakes, and full of edge cases, exactly the conditions where a polished demo falls apart.

That was the point. As the session went on, we introduced new constraints and data anomalies on purpose, mirroring how real AI and ML projects actually unfold. The interesting learning is not "the model works." It is what you do when the data is wrong, the format changes, or the answer has to be auditable.

Why we teach it hands-on

You do not learn AWS AI from slides. You learn it by wiring services together, hitting a wall, and working out why. Putting engineers, data scientists, and business folks on the same problem also surfaces something a lecture cannot: the real bottleneck in AI projects is rarely the model. It is the data plumbing, the collaboration across roles, and the willingness to adapt when conditions change.

Want this for your team?

Teaching this is some of our favorite work, and it is the same thing we do for clients: get teams genuinely capable on AWS AI, Bedrock, Knowledge Bases, and the data work that makes them useful, on real problems instead of toy examples.

If your team wants a hands-on session like this, or help putting Bedrock and Knowledge Bases into production, let's talk.