From a Book, to a CRM, to an AI: The Accidental Path to Building A Coach App

By Andrei Roman

Principal Architect, The Foundry

This is the story of how I started using AI to write code and learn new things.

Step 1: The Tutorial (Following the Map)

I've spent most of my career working in safety-critical software. But the world of modern full-stack web development was a new territory.

And my personal projects were taking me there. I had been using Excel as a personal tracking tool for a while (when JIRA is too much...), and also Apple Notes. But I wanted an integrated solution instead of copy pasting content between tools.

A trusted friend and fellow engineer - a member of my "Council" - gave me my first mission:"Go through 'Full Stack Development with Spring Boot and React.' Build the CRUD app. Master the fundamentals."

So I did. I treated it like a professional assignment. I finished the book, and I had a working, but generic, application. I had followed the map. I had the basic skills.

Step 2: The Personal Project (Drawing My Own Map)

A tutorial is not a product. To truly master the craft, I needed to solve a real problem. My own was chaos. My personal projects were a mess of spreadsheets, notes, and a complete lack of a coherent status overview.

By then I had started using ChatGPT for my personal projects so I took the next step - to build the project tracking tool of my dreams. It was an ambitious mix: an Excel-like interface for tasks, an Apple Notes-style organization for projects, and a way to automatically generate status updates. I was obsessed. I coded on it every spare moment, even while visiting one of my wife's clients in the countryside.

Step 3: The Accidental Client (The First Real Test)

The small business owner we were visiting saw the project tracker over my shoulder. He saw the structure, the organization.

"My team is drowning in emails and spreadsheets," he told me. "Can you build something like that, but for managing our customers?"

He wasn't asking for a project tracker; he was asking for a custom CRM. And he was willing to pay for it. My personal learning project had just become my first real-world business application.

After meeting their team and gathering their first requirements, I dove in. The first version worked, but as the features grew, I hit a wall. It wasn't that it couldn't scale; the problem was that it was becoming increasingly hard to change. Business logic was tangled in services and controllers. A simple modification in one place would cause a cascade of bugs elsewhere. The system was accumulating "technical debt" at an alarming rate. This is where my "Council" member intervened again, introducing me to the principles of Clean Architecture - a revelation that became the new foundation for all my future work. I knew clean code, but this was extending the concepts to entire projects.

Step 4: The AI Catalyst (Solving a Problem at Home)

With the CRM delivered, I had a new, robust architectural framework. But the next leap came from my own home. My wife, a freelancer managing multiple WordPress sites, was frustrated with generic AI tools.

"ChatGPT is useless for my workflow," she explained. "Every time I want to generate content for a client, I have to re-explain the entire context: who the client is, their audience, the tone of voice, the specific page I'm working on."

She needed an AI that wasn't just smart; she needed an AI that was context-aware.

That became my next obsession. I built her a tool. A simple interface where she could select a site and a page, and when she opened the AI chat, the tool would automatically inject the entire context into the prompt. It was a workflow improvement. In building it, I learned the fundamentals - first working with the REST API, then defining tools (have the AI retrieve a page and analyze it for example).

The Final Convergence: From Tools to Mission

I now had a complete, battle-tested arsenal:

• A full-stack development framework built on Clean Architecture.

• Experience with payments and security.

• A deep, practical understanding of building context-aware AI with RAG.

I was an engineer with a powerful set of solutions, looking for my true problem to solve.

And when the guys were discussing their marriage issues, it all clicked into place.

The 20-year-old dream resurfaced:"I wish I had an application to keep me on track, to be the brutally honest coach in my pocket." Guided by a very specific set of principles.

That is how my Coach app was born (currently offline - I completely refactored it on top of FRAKTAG and using it for testing - it is very useful compared to simple RAG because it can apply a hierarchy of rules - this is a must for coaching).

Welcome to the forge.