How Do Solo Consultants Build an AI Knowledge Base?

Why do solo consultants need a self-improving knowledge base?

As a solo consultant, your expertise is your product. But that expertise lives in scattered places — old proposal decks, client debrief notes buried in email, saved articles you'll never re-read, book highlights trapped in Kindle. Every time you start a new engagement, you're rebuilding context from memory instead of drawing on everything you've ever learned.

The Karpathy Self-Improving AI Knowledge Base solves this by making an AI your librarian. You dump all your material into a Raw folder — unorganized, unsorted — and the AI compiles it into a structured Wiki with cross-linked topics, an index, and discoverable connections between ideas you didn't even know were related.

How do you set up the knowledge base for consulting work?

Start by creating a domain folder like `consulting-kb` inside a top-level second brain folder. Add three subfolders: Raw, Wiki, and Outputs. Write a Claude MD schema file specifying your focus areas — for example, "client stakeholder management," "pricing and scoping," and "delivery methodology."

Then dump everything you have: past proposals (converted to markdown), meeting notes, saved articles on consulting strategy, book highlights, even voice memo transcripts. Don't organize any of it. Drop it into Raw and move on.

Prompt your AI to read everything in Raw and compile a Wiki following the Claude MD rules. The AI creates one markdown file per major topic, builds an index, and links related concepts. This initial build takes about 30 minutes of AI processing time — you can do other work while it runs.

How does the knowledge base help you win and deliver client work?

Once your Wiki is built, you can query it like a research assistant who has read everything you've ever saved. Ask: "What does my knowledge base say about handling a client who wants to expand scope without increasing budget?" The AI pulls relevant Wiki entries, cites your own past notes and sources, and generates a report saved to Outputs.

The compounding loop is where the real value lives. Every question you ask and every answer generated gets saved back into the system. After three months of regular use, your knowledge base contains not just your original material but dozens of synthesized reports — each one making the next answer richer and more specific.

Run a monthly health check to keep the system sharp. The AI audits for contradictions, coverage gaps, and stale content. It might flag that your Wiki has extensive content on pricing but nothing on subcontractor management — then propose and draft new articles to fill the gap.

What makes this different from keeping notes in Notion?

Notion requires you to be the librarian. You create databases, tag entries, build views, and maintain links manually. The moment you fall behind on organization (which every consultant does), the system stops working. The Karpathy method eliminates this entirely — the AI handles all organization, and the system actually improves when you neglect tidiness because Raw is designed to be messy.

The compounding effect is the other key difference. Notion notes are static. This knowledge base grows smarter with every interaction, surfacing connections and filling gaps automatically through monthly health checks.

Next step

Create your first domain folder today. Spend 15 minutes dumping your most recent client project notes into Raw. Write a basic Claude MD with three focus themes relevant to your practice. Run the Wiki build prompt and ask your first question. The system is deliberately thin on day one — commit to 30 days of regular use and watch it compound into something no competitor can replicate.