// Building with AI · Live

Self-Improving Knowledge Base

A notebook that reads what you save, keeps only what holds up, and points at the problems worth building for.

Everyone who follows AI has a graveyard of saved articles. This project turns that pile into a knowledge base that works for you. Each note is filtered through one question: what real problem is this, and could software solve it for recurring money?

We built it with two models. One built the engine. The second audited it, then added the scoring and a skeptic pass that tries to refute every note before it counts. The case study below shows the method, and what the first run actually produced.

// Who it's for

▸ People with a pile of saved articles and newsletters that never turn into anything
▸ Builders looking for a grounded product idea rather than a hunch

// Who it isn't for

▸ Anyone who wants a search box. This makes notes and argues with them; it doesn't answer questions about your files.

// Proof it works

50 notes distilled so far, across five types: problems, revenue patterns, build techniques, distribution tactics and opportunitiesas of 2026-09-28
The first run surfaced a concrete product idea, written up in the case study below

// Build it yourself, build it with us, or have it built

Build it yourself

Starter prompts

Free

You can already build. Copy these and go. No email needed.

Distil one saved article into an evergreen note

Paste one article, newsletter or transcript. You get a short note: what it is, the real problem behind it, whether there's a path to recurring revenue, and what you'd need to find out first.

You are the distiller for my personal knowledge base. I save articles, newsletters and transcripts. Your job is to turn one saved item into a single evergreen note that I will still find useful in a year, filtered through one lens: what real problem does this reveal, and how could software solve it for recurring revenue?

Write the note in markdown with exactly these sections:

## What it is
2-3 sentences in your own words.

## Why it matters (the problem)
Who has this problem, how painful it is, and the evidence for that in the source. If the source doesn't show a real problem, say so plainly.

## Path to revenue
How a product or service could make recurring money from it. If there isn't a credible path, write "none yet" and say why. Do not invent prices, market sizes or customer numbers.

## Open questions
2-4 things you would need to find out before building anything.

## Sources
The source title and URL exactly as given.

Rules:
- Every claim must come from the source below. If the source doesn't support it, leave it out.
- No hype words ("revolutionary", "game-changing"). Plain, specific language.
- Keep the whole note under 350 words.

Source title: {{TITLE}}
Source URL: {{URL}}
Source text:
{{PASTE THE ARTICLE OR TRANSCRIPT}}

Tested on Claude Opus 5.5 on 28 September 2026 with our own article on NVIDIA's SoL-Pi. The note kept the article's cost figures and caveats, cited only the one source URL, and answered the revenue question with "unclear, and possibly none yet" rather than inventing a market.

Build it yourself

Pro pack

$29

Builders who run Claude Code and want the full prompt set and note schema, ready to use.

The full prompt set: capture, distill, review, prove, and turn an opportunity into a spec
The note schema and folder layout
Setup for Claude Code, including capturing member-only articles you've paid for into your private notes

Planned contents. Details may change before release.

Coming soon

This also subscribes you to our free weekly newsletter. Unsubscribe any time.

Build it with us

Guided build

$49

You can't build yet. Every step, every account, every command. Pro pack included.

Everything in the pro pack
Step by step from an empty folder to your first weekly distill
How the two-model method works, and how to score the knowledge base so you can tell it's getting better