AI knowledge bases
Your company's knowledge, on call. Not stuck in someone's head.
Every company runs on what its people know: how jobs get priced, why that client gets special handling, where the file is. I turn that knowledge into an AI knowledge base your whole team can ask, inside the tools they already use.
The problem
Your most valuable asset has no backup.
Knowledge is the one thing every company depends on and almost none of them manage.
01
It walks out the door
When a senior person retires or quits, their judgment leaves with them. The next hire starts from zero.
02
It's buried
SOPs, drives, old proposals and call recordings. It's all technically there, and nobody can find it when it matters.
03
Your AI doesn't know it
ChatGPT gives generic answers about your business. Worse, your team pastes company data into tools you don't control.
How I build it
Extract. Structure. Deliver. Measure.
01
Extract
Out of heads and files
Interviews with the people who know, plus the SOPs, documents and recordings you already have. I use the same voice extraction process I've run with experts for years.
02
Structure
Skills, not a document dump
Knowledge gets broken into small, specific skills: how to price this, what to check on that, who to call when. Each one is reviewable by the person who owns it.
03
Deliver
Inside the tools people use
Your team asks in Claude, ChatGPT, Copilot or Slack. Field crews ask by text. Nobody has to learn a new system or remember where the wiki is.
04
Measure
See the gaps
You see what people ask and where the answers are thin. Those gaps become the next thing to capture, so the knowledge base gets better every month.
Powered by Skill Refinery
I built the platform for exactly this.
Skill Refinery is my AI knowledge delivery platform. It turns expert knowledge into structured skills and delivers them to the AI tools your team already uses. Your knowledge base runs on it, so you're not paying me to rebuild the plumbing.
Owned by you
Your knowledge stays yours. Source files aren't used to train AI models.
Delivered through MCP
The open standard that plugs knowledge straight into Claude, ChatGPT and other AI tools.
Built to learn
Usage and gap signals show what your team actually needs, not what someone guessed they'd need.
Where it pays
Every team runs on knowledge. Start with one.
Most clients start with the role where knowledge is most expensive to lose, then expand.
Onboarding
New hires ask the knowledge base instead of interrupting your best people. Ramp time drops from months to weeks.
Field and operations
Crews and techs get the spec, the procedure or the job history from their phone, on site.
Sales enablement
Every rep gets your top performer's answers to pricing, objections and scoping questions.
Compliance and quality
The right procedure, every time, with a record of what was asked and answered.
Succession and M&A
Capture what the founder and key people know before a sale, a retirement or an acquisition. Buyers pay for knowledge that stays.
Your expertise as a product
If you sell a methodology, clients can use it through AI between sessions. Skill Refinery handles delivery and billing.
Running now
Knowledge systems in production.
Commercial fire protection contractor
214K files
A field-ops assistant that knows every job file
An assistant in Slack trained on the company's 214,000 files and SOPs. It pulls project documents by job number, answers crews in English or Spanish, sends QC texts to the field and logs the replies back to the job.
Read the case →Multi-clinic physical therapy group
Every clinic
Authorizations and visit caps that track themselves
An operations system that stays in sync with the EMR, flags expiring authorizations and maxed visits before the appointment, and puts cases back on the schedule as soon as the paperwork arrives.
Read the case →Questions
What people ask about knowledge bases.
Isn't this just a chatbot on our documents?
No. A chatbot on a document pile gives you confident answers from outdated files. I extract and structure the knowledge first, have the people who own it review it, and then deliver it. The structure is the difference.
Will our data be used to train someone's AI?
No. Your knowledge stays yours, and source files aren't used to train AI models.
Which AI tools does it work with?
Claude, ChatGPT, Microsoft Copilot and Slack, plus text for people in the field. If your team already lives in one of them, that's where it goes.
Our documents are a mess. Is that a problem?
That's normal. Most of the valuable knowledge isn't in documents anyway; it's in people's heads. Cleaning up what exists is part of the build.
How long does it take?
A first knowledge base covering one team or one role is usually live in weeks. It grows from there.
Next step
Which knowledge would hurt most to lose?
That's where we start. Thirty minutes with me and you'll know what to capture first.