Putting the Pieces Together

The Story of Keryk_AI: Part 4

Originally published on LinkedIn, May 1 2025

Over the last few posts, I’ve been sharing how a mix of curiosity, personal projects, and real-world conversations led me to think differently about AI.

It started with my son, Wade, and a few tech experiments during a tough job market. Then came long daily walks, where I discovered how voice AI could help me think faster, clarify my ideas, and turn raw thoughts into real output. A little while later, a phone call with my old friend and college roommate reframed how I thought about the kind of work I wanted to do next.

And somewhere along the way, all those threads started to connect.

The realization

At the core of all this was a single, simple idea:

The biggest untapped resource in most organizations is what their people know.

Not the raw data in their systems. Not their documentation. What’s in their heads.

And most of the time, that knowledge stays locked away, either because no one has time to extract it, or because there’s no good system for turning it into something useful, repeatable, and scalable.

Once I started testing voice-based AI to help people articulate what they knew, it clicked. The value wasn’t in getting AI to generate content, it was in helping people surface and structure the thinking they already had. And once you’ve got that, you can use it in all kinds of powerful ways.

That’s when I started looking at the bigger picture:

• How could we capture expertise faster, with less friction?
• How could we structure that knowledge so both humans and AI could use it?
• How could we apply that knowledge through agents, documents, dashboards, and decisions?
• And how could we measure the business impact?

The Turning Point

The real turning point came during the catch-up call with Dave.

We were talking about career paths for folks like us; people with a lot of experience, still wanting to build and contribute, but not necessarily in a traditional staff role. Dave mentioned hourly consulting, which is a route a lot of people take. It’s simple, familiar, and can be rewarding.

But I found myself explaining why that model wasn’t the right fit for me. It’s not just the paperwork. It’s the fact that hourly work is inherently limited. You only have so many hours in a day, and only so much people are willing to pay per hour.

That conversation got me thinking:

What if there was a way to to multiply your impact without multiplying your hours?

And that question changed everything.

From hourly to exponential

 I started testing whether AI could help not just with thinking and writing, but with real work. Could it help me analyze data? Build early prototypes? Model financial outcomes? Spot process gaps?

I began experimenting with agentic frameworks and process flows,  breaking problems down into steps and asking, “How could AI assist with this part?”

Everywhere I looked, there were opportunities to speed things up, not by replacing people, but by letting AI handle the tedious parts so humans could focus on the judgment calls. But it wasn’t just about speed. It was about scale, doing more with the same amount of effort, and delivering outcomes that didn’t rely on trading time for money.

Of course, it wasn’t perfect. Generative AI and autonomous agents are non-deterministic, so the same input doesn’t always give the same output. So I followed the research on how to build structure around it:

• Clear steps and flows
• Human checkpoints
• Guardrails to keep things stable and explainable

That’s the art: not just what AI can do, but how to design systems that use it well.

Building the infrastructure

As I built out the pipeline, I realized something else: capturing knowledge is only part of the problem. You also need to store and retrieve it in ways that preserve meaning, relationships, and context.

That’s when I dove deep into ontologies, graph databases, and vector databases to provide a more comprehensive way to capture knowledge. I combined them into a hybrid RAG (Retrieval-Augmented Generation) approach. Now when I asked a question of the system, it could return not just the right document, but the right idea, the right explanation, or even the right next step. Conceptually aware, contextually accurate, and highly usable.

That’s when I knew I wasn’t just building tools—I was building a system.

Introducing Keryk AI

All of this came together in what is now Keryk AI

Keryk is focused on helping organizations unlock their internal expertise and turn it into something scalable, measurable, and usable.

At the core is our architecture: the PICK stack ( Pipeline for Integration of Contextual Knowledge.) It’s designed to:

1. Capture subject matter expertise through voice-guided AI
2. Structure that knowledge using ontologies, graphs, and vectors
3. Apply it through AI-powered agents, chatbots, documents, and dashboards
4. Align everything to business goals through measurable ROI

Eating my own dog food (or drinking my own champagne)

This system isn’t just built for business, it’s built from business.

I’ve spent my career working in the trenches of process design, automation, operations, financial analysis, and technology adoption. All of that experience is no embedded into PICK.  The first Keryk “test subject” was me.

Every part of the PICK stack reflects real-world principles I’ve used for decades:

• Process optimization
• Productivity modeling
• Business case development
• Systems thinking
• ROI-driven decision-making

That’s why I believe this is something different.

Other companies could copy the same stack, the same tools, even the same interfaces—and still not get the same results. Because what makes it work is the experience behind how it’s applied and how the components are directed—through the software code, AI prompts, and agent design that all contain embedded versions of my specific experience. 

On top of that, it’s designed to learn from itself. As the system captures ideas, builds agents, and supports real-world projects, it generates data that can be analyzed to uncover new patterns, spot better ways of doing things, and surface emerging best practices. Over time, it gets smarter, not just at execution, but at refinement.

How Keryk Works

Keryk is designed to meet companies where they are, with fast-start services that create momentum, not overhead.

We start with a short, focused engagement called Jumpstart—usually 3 to 4 weeks. It includes strategy sessions, a human-assisted, AI voice-guided SME interview, followed by analysis of high-value use cases and a working demo of an AI-powered app or agent built on your own internal knowledge.

 The goal is to show what’s possible while also starting to uncover the areas with the highest ROI potential and greatest business impact, so future efforts are targeted where they’ll deliver the most value

From there, companies can choose to engage for a Knowledge Activation project, to dig deeper and explore more complex use cases. Later, they can extend with follow-on projects or subscribe to an annual Advisory Service model that includes ROI tracking, knowledge base updates, and development of new use cases over time.

Everything we build is tailored to your business, not based on pre-trained data or generic templates. And every part of the process is backed by a system designed to scale what your team already knows without disrupting how they work. What’s next

Now that the story’s out there, I’m planning to mix things up a bit with future posts.

Some will be Keryk-specific to share insights from what we’re building and what we’re learning. Others will be about AI in general, to share some cool tools, practical tips, or things I’ve found helpful.

Thanks for following along so far. If you’ve made it to the end of this series, I appreciate you.

There’s a lot more to come and I’m excited to share it.