How to Build an AI Knowledge Base as a Solo Consultant
For solo consultants · Based on Karpathy Self-Improving AI Knowledge Base
// TL;DR
Solo consultants can build a self-improving AI knowledge base to stop losing client insights across Notion, downloads folders, and email. You dump saved blog posts, book highlights, and past client debrief notes into a Raw folder, and an AI librarian compiles a cross-linked Wiki on your practice area — negotiation, strategy, whatever you specialise in. Then you query it before proposals and calls, save the answers back, and run a monthly health check that flags gaps in your expertise. Over a few months, it becomes a proprietary advisory asset no competitor can replicate.
Why do solo consultants need an AI knowledge base?
As a solo consultant, your entire value is the depth and recall of what you know. But that knowledge is scattered — client debrief notes in Notion, book highlights in Kindle, saved blog posts in a browser tab graveyard, meeting transcripts in your downloads folder. You re-search the same things constantly and lose hard-won insights from past engagements. The Karpathy self-improving AI knowledge base fixes this by making an AI your librarian: you dump everything into one Raw folder, and the AI organises, links, and surfaces it on demand.
How do I set it up for my consulting practice?
Create a top-level second brain folder, then a domain folder — say `negotiation-kb`. Inside it, add three subfolders: Raw, Wiki, and Outputs, plus a Claude MD schema file at the root. In the Claude MD, define three themed focus areas that match your practice, for example: persuasion principles, difficult conversation frameworks, and commercial negotiation tactics. These themes tune the AI to deepen coverage where you actually advise clients.
Then dump everything into Raw. Export your Notion pages, save your blog posts as markdown using the Obsidian web clipper, and drop in your past client debrief meeting notes. Do not organise or rename anything — Raw is a junk drawer by design. This should take 10–15 minutes.
How do I turn raw notes into usable insight?
Point the AI at the folder and give it one prompt: read everything in Raw and compile a Wiki following the Claude MD, creating the index first, then one file per major topic with cross-links. Walk away for about 30 minutes. What comes back is a cross-linked Wiki of your entire practice knowledge — topic pages with summaries and discovered connections between ideas you'd never manually drawn.
Now query it in real work. Before a tricky call, ask: 'What does my knowledge base say about handling a client who disputes scope mid-project?' The AI pulls the relevant Wiki entries and answers with your own sources cited. Save that report to Outputs. Then ask for your three biggest knowledge gaps and save that too — this is the compounding loop that makes tomorrow's answer sharper than today's.
How does the monthly health check help my practice grow?
At month end, run the seven-stage health check. For a consultant, it typically surfaces things like: your Wiki has plenty on verbal negotiation but nothing on written negotiation, or a claim in one article contradicts a book highlight in another. The AI proposes new article candidates and, in interactive mode, drafts them for you to review and ingest. Over three to four months of this, you accumulate a proprietary advisory knowledge base — the exact edge that lets you charge more than consultants relying on generic frameworks.
Remember the discipline: never edit the Wiki by hand, always save outputs back, and keep the Change Log updated so the AI knows what's new. Day one is basic. By day 100 you have something no competitor can copy.
Next step: Create your second brain folder today, spin up one domain folder for your core practice area, and spend 15 minutes dumping your existing client notes and book highlights into Raw before your next Wiki build.
// FREQUENTLY ASKED QUESTIONS
What consulting material should I put in Raw first?
Start with your highest-value scattered material: past client debrief notes, book highlights on your practice area, saved blog posts, and meeting transcripts. Export Notion pages to markdown and use the Obsidian web clipper for web articles. Don't curate or organise — dump it all in unedited so the AI can find connections across everything you've accumulated.
Can this help me prepare for specific client calls?
Yes. Before a call, start a new session, point the AI at your knowledge base, and ask a targeted question like how to handle a mid-project scope dispute. The AI pulls relevant Wiki entries and answers with your own sources. Save that report to Outputs so future call prep gets faster and sharper each time.
How do I keep client-specific information organised?
Use themed focus areas in your Claude MD to structure your domain — for example persuasion principles, difficult conversations, and commercial tactics. For distinct practice areas, create separate knowledge base folders under one second brain container. Keep confidential client details in a separate base and stagger health checks across different days to manage session credits.