I Started Organizing a Decade of Knowledge with AI — Why and Where to Begin

Not long ago, I ran into an uncomfortable question.

“Where is everything I’ve built up over the past ten years?”

Working in OTT streaming and content security, I’d read hundreds of articles, shipped many projects, and solved a lot of problems. But most of that knowledge was sitting in internal docs, tech blog posts, read-later bookmarks, and my own head, with nothing linking it together.

What AI Can’t Fill In

When AI tools started flooding the market, my first reaction was simple: productivity boost. They summarize, draft, write code.

Then I hit a realization.

There’s one thing AI can’t fill in: the reflection that comes from your own experience.

Ask ChatGPT what the most common mistakes are in DRM implementation on an OTT platform and you’ll get a plausible answer. But that answer doesn’t carry the weight of a specific failure I’ve lived through, the lesson from a particular conversation with a customer, or the intuition those experiences built into me. That part I have to surface myself.

Starting to Build an LLM-Wiki

Over the past few days I’ve been building a personal knowledge management system around the LLM-Wiki approach.

This isn’t a concept I invented. It was developed and shared by Andrej Karpathy, an AI researcher who came out of OpenAI. (See the original) The original is personal-note-centric; I’ve added domain-specific structure and an AI-agent collaboration layer on top.

The concept is simple:

• I collect articles, papers, and experiences as raw sources.
• An AI agent (Claude Code) analyzes them and extracts the key points.
• I curate the output into wiki pages: concept definitions, entities (people, companies, technologies), and synthesized insights.
• As each new source comes in, the AI agent cross-references it against existing wiki pages and surfaces connections. Each session, the knowledge compounds.

Process is what matters, not the tools.

AI handles the mechanical work — summarizing, indexing, cross-referencing. I focus on judgment and reflection. When you can’t carve out long stretches of time for this, splitting the roles makes limited time go much further.

A Few Weeks In

As of when I’m writing this, the system contains:

• 5 domains (OTT/streaming, DRM/content security, cloud/SaaS, AI/LLM, solopreneurship)
• 24 wiki pages: concept definitions, entities, synthesized insights
• 49 sources: read, analyzed, and registered in the system (I’ll call this process “ingesting” throughout this series)

The number that matters most isn’t on that list.

Last week someone asked me how to approach key rotation strategy in a multi-DRM environment. I didn’t scan my memory. I opened the wiki, pulled up the structured entry, and walked them through my thinking far more clearly than I could have before. The difference between reaching for the wiki and searching your memory seems small. It compounds.

Why I’m Documenting This in Public

This isn’t just my story.

I see plenty of people around me in the same situation: deep expertise that hasn’t been organized. An interest in using AI, but no clear sense of where to connect it.

Two reasons I’m writing this publicly.

First, accountability. Writing forces more precise thinking.

Second, to be useful to someone on a similar path. The trial and error I’ve gone through can save someone else time.

What’s Coming in This Series

Topics ahead:

• How I select and ingest sources
• The actual workflow for collaborating with an AI agent
• The moment disconnected sources start connecting
• How a knowledge wiki becomes raw material for writing and eventually an ebook

At the end, I plan to cover packaging the methodology into something others can use.

If you have deep experience in a field but haven’t worked out how to organize and use it, I’d love to have you follow along.

Comments and DMs are always welcome.