How Do Solo Consultants Build an AI Knowledge Base?
For Solo consultants and freelancers · Based on Karpathy Self-Improving AI Knowledge Base
// TL;DR
Solo consultants accumulate client insights, frameworks, articles, and meeting notes across dozens of projects but rarely have time to organise them. The Karpathy Self-Improving AI Knowledge Base lets you dump everything into a Raw folder and have Claude organise it into a searchable, cross-linked Wiki. Over time, every client question you answer feeds back into the system, making your next proposal sharper, your recommendations more grounded, and your expertise compounding. Set it up once, use it daily, and by month three you have a proprietary asset no competitor can replicate.
Why Do Solo Consultants Need an AI-Managed Knowledge Base?
Most solo consultants have years of accumulated knowledge scattered across Google Docs, Notion, email threads, and their downloads folder. They know the insights are there somewhere, but finding and synthesising them under time pressure is almost impossible.
The Karpathy Self-Improving AI Knowledge Base solves this by making the AI your librarian. You never organise anything. You dump articles, client debrief notes, book highlights, proposal drafts, and framework sketches into a Raw folder. Claude reads it all, builds a cross-linked Wiki, and answers your questions by citing your own collected knowledge.
For a consultant, this means faster proposal writing, better-informed client recommendations, and a growing competitive moat.
How Do You Set Up the Knowledge Base for Consulting Work?
Start by creating a folder structure: a top-level second brain folder containing a domain folder like `consulting-kb`. Inside that, create Raw, Wiki, and Outputs subfolders plus a Claude MD schema file.
In the Claude MD, define three to five themed focus areas relevant to your practice. A strategy consultant might use: competitive analysis frameworks, stakeholder alignment techniques, and pricing strategy. A marketing consultant might use: channel attribution, brand positioning, and content strategy ROI.
Then dump everything you have. Past client reports, saved articles, conference notes, book highlights — all into Raw as markdown files. Use the Obsidian web clipper to convert saved web pages in one click. Don't organise anything. Messy is by design.
Prompt Claude: Read everything in Raw and compile a Wiki following the rules in your Claude MD. Create the index first, then one file per major topic with cross-links. Let it run for 30 minutes.
How Does the System Improve Your Client Work Over Time?
The magic is the compounding loop. Every time you query the knowledge base — "What does my knowledge say about handling scope creep in enterprise deals?" — the answer gets saved to Outputs. Every month, the health check finds gaps: maybe you have extensive notes on pricing but nothing on procurement negotiation.
Those gap reports tell you exactly what to learn next. Commission content, save articles, or add notes from your next client engagement. The system gets smarter with each cycle.
By month three, you can generate client-ready briefings in minutes that would have taken hours of manual research. By month six, the knowledge base surfaces non-obvious connections between past projects that inform new ones. This is the kind of proprietary advantage that justifies premium rates.
What's the Monthly Maintenance Like for a Busy Consultant?
Minimal. Dumping new content into Raw takes seconds — just drag files or paste into Claude and ask it to save to Raw. The monthly health check takes 15 to 30 minutes of interactive review. The AI does the heavy lifting: finding contradictions, flagging stale content, proposing new articles. You just approve or reject its suggestions.
Schedule it on the first Monday of each month. Treat it like a systems check, not a project. The compound returns are worth far more than the 30 minutes invested.
Ready to start? Create your folder structure, write your Claude MD with your consulting focus areas, and dump your first 20 pieces of content into Raw today.
// FREQUENTLY ASKED QUESTIONS
What kind of consulting content should I put in the Raw folder?
Everything related to your domain: client debrief notes, proposal drafts, saved articles, book highlights, conference talk notes, framework diagrams (as text descriptions), competitor analyses, and meeting transcripts. Don't filter or organise — the AI handles that. The more varied your inputs, the richer the connections the Wiki will surface.
Can I use the knowledge base to help write client proposals?
Yes. Query the system with the client's specific challenge and the AI will generate a briefing citing your own collected knowledge, past frameworks, and relevant insights. Save that briefing to Outputs. Over time, your proposal inputs become richer because each past engagement's learnings are already in the system. This cuts proposal writing time dramatically.
How do I keep client information confidential in the knowledge base?
The entire system runs locally on your machine using markdown files and Claude's file-system access. Nothing is uploaded to a shared cloud unless you put it there. Anonymise client names in Raw files if needed. The knowledge base is as secure as your local file system. For extra caution, store the second brain folder in an encrypted volume.