大齐 daqi

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

Software Engineering Philosophy & Development Protocol

1. Global Integrity & Topology

System Awareness: Before implementation, map the dependency graph and system topology. Ensure local changes preserve global invariants and do not trigger "Shotgun Surgery."
Orthogonality: Design for independence. Ensure that changes in one module do not leak side effects into others. Minimize coupling and maximize cohesion.

2. Intent & Abstraction Hierarchy

Intent-Revealing Design: Prioritize human readability and intent over machine cleverness. Use naming that explains "Why" rather than "How."
Single Level of Abstraction (SLA): Adhere strictly to the Single Responsibility Principle (SRP). Each function must operate at a consistent level of abstraction and have exactly one reason to change.

Styling native elements

Native HTML controls are a challenge to style. You can style any element in the web platform that uses Shadow DOM with a pseudo element ::pseudo-element or the /deep/ path selector.

video::webkit-media-controls-timeline {
  background-color: lime;
}

video /deep/ input[type=range] {

	"""
	The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp

	const ignoreErrors = [
	/^No error$/,
	/__show__deepen/,
	/_avast_submit/,
	/Access is denied/,
	/anonymous function: captureException/,
	/Blocked a frame with origin/,
	/can't redefine non-configurable property "userAgent"/,
	/change_ua/,
	/console is not defined/,