06.10.2025: Building a Smarter Way to Search My Archive

June 10, 2025, was the spark day—the day the idea for AutonomousMatt transitioned from a thought experiment into something that demanded execution. It was raw, impulsive, speculative. No files, no models, no logic trees. Just a browser window, a blank text editor, and a single driving question:

What if GPT could speak from my actual body of work, not just mimic my tone?

Here's everything that happened on the day this project was born.

💥 The Idea Emerges

1. The Core Concept Took Shape

The idea wasn’t to build a chatbot. It was to build a searchable, conversational archive of my intellectual and creative life—a personal knowledge engine trained not on general data, but on my articles, essays, presentations, reviews, and reflections.

A model that could say what I’ve said—faithfully—and explain what I meant.

It began with a sketch in my notes app:

vbnet

CopyEdit

Ask me anything. Get a response drawn from decades of my writing. Let the archive speak.

🧭 Guiding Principles Were Drafted

I started writing down the rules the system should follow:

• Everything must be sourced — no hallucinations, no vague summaries

• The tone must remain mine — but evolved, clarified, sharpened

• Answers should feel like essays, not blurbs

• No “training” on the model side—context only, from real files

• Text files, not a CMS—lightweight, durable, portable

This wasn’t about AI automation. It was about human authorship at scale.

🧱 Technical Scaffolding Was Imagined

2. Outlined the Basic App Flow

I mapped the MVP structure in a single document:

1. User asks a question

2. JS captures and sanitizes the prompt

3. The app matches the prompt to a keyword

4. The keyword maps to a .txt file

5. That file is passed as GPT context

6. GPT returns a bespoke response, framed by the original work

7. Response is displayed with source citation

At this point, it was all theoretical. No HTML. No JS. Just paper sketches and mental models.

3. Picked the File Format: .txt

I decided early that I wouldn’t build this on a database or CMS. Each piece of content would live in a plaintext .txt file—hand-editable, portable, archivable. The system would grow by curating files, not coding them.

🌐 Domain + Identity Decisions

4. Named the Project: AutonomousMatt

The name came fast and stuck. It carried layers:

• Autonomous → independent, self-guided, responsive

• Matt → personal, authored, specific

• Together → an AI interface to me, with boundaries

I immediately reserved the domain: autonomousmatt.com

5. Drafted First Copy for the Site

I sketched the early tagline:

Effortlessly access the knowledge in hundreds of articles, presentations, papers and reviews from thirty years of curated product management, design, marketing, academic and creative projects. Built from the career archives of creative product strategist Matt Shadbolt. Ask me anything.

This language would evolve later, but the core promise was clear: a dialogic front-end to an intellectual history.

🧠 Conceptual Breakthroughs

6. Defined the System’s Voice

I asked: How should this thing sound?

Answer:

• First-person, reflective

• Knowledgeable, but personal

• Essay-like, not transactional

• Willing to say “I wrote about that here…”

In short, not ChatGPT pretending to be Matt—but Matt, with help from GPT, reaching back through time.

7. Acknowledged the Limitations (and Potential)

Even at this early stage, I knew:

• GPT would need careful prompting to stay accurate

• Some pieces would need to be rewritten or restructured

• The system would always have gaps—and that’s okay

• This project wasn’t about completeness. It was about presence

🛠️ What Didn’t Exist Yet

• No front-end

• No file routing

• No AI integration

• No keyword logic

• No formatted archive content

But the mental framework was built. And once that exists, the rest is execution.

🔭 What the Day Set in Motion

June 10 set a creative fire. Over the next four days, that idea would move from sketch to working prototype:

• June 11: Assembled content and created the schema

• June 12: Finalized file structure and metadata format

• June 13–14: Deployed the engine, keyword map, and archive logic

But it started here—with one question, typed into a doc:

What if the archive could answer?

🧠 Final Thought

June 10 was ignition. There was no product yet. No code. Just intent.

But that’s always how these things begin—not with a feature list, but with a feeling.

There’s something in here that still needs to speak.

So I built a system to let it.

—Matt

Explore it now at autonomousmatt.com
Or just ask me something I once knew.

---