Kihoon Lee (Kilee) Likilee

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.

Git Compression of Blobs and Packfiles.

Many users of Git are curious about the lack of delta compression at the object (blob) level when commits are first written. This efficiency is saved until the pack file is written. Loose objects are written in compressed, but non-delta format at the time of each commit.

A simple run though of a commit sequence with only the smallest change to the image (in uncompressed TIFF format to amplify the observable behavior) aids the understanding of this deferred and different approach efficiency.

The command sequence:

Create the repo:

	import 'package:flutter/material.dart';

	void main() {
	runApp(const MyApp());
	}

	class MyApp extends StatelessWidget {
	const MyApp({super.key});

	@override

	#/bin/bash

	BANNER="$(
	cat <<-EOF
	# **************************************************************************** #
	# #
	# ::: :::::::: #
	# link_all_cache_dirs.sh :+: :+: :+: #
	# +:+ +:+ +:+ #
	# By: jaeskim <[email protected]> +#+ +:+ +#+ #

	let cache = new Map();
	let pending = new Map();

	function fetchTextSync(url) {
	if (cache.has(url)) {
	return cache.get(url);
	}
	if (pending.has(url)) {
	throw pending.get(url);
	}