dh dhfromkorea

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.

	desc "Default deploy task"
	task :deploy do
	# Check if preview posts exist, which should not be published
	if File.exists?(".preview-mode")
	puts "## Found posts in preview mode, regenerating files ..."
	File.delete(".preview-mode")
	Rake::Task[:generate].execute
	end
	Rake::Task[:copydot].invoke(source_dir, public_dir)

	# this model is the most naive one
	# find the occurences of the keywords listed and consider them to be matched.
	# TODO will be to apply a Bag of Words model or some sort and assign credit score

	# read raw caption files
	caption_data, metadata = next(load_caption_files(path_to_caption_file))
	# combine captions by 10 seconds (i.e. make time blocks)
	X = split_caption_to_X(caption_data)

	keywords = ['caption', 'type=story', 'type=commercial']

	fp = FingerPrint(path_to_dejavu_config_file)
	timestamp_pb = 20000 # program boundary timestamp in secs from
	PAD = 10 # temporal padding to apply to the timestamp

	# slice an audio segement off the video
	fp.slice_audio_segment(path_to_full_video_or_audio, pb - PAD, pb + PAD, "mp4")

	# extract the fingerprint of the segment and save it to db
	# db's configured in the .config file
	# the path for the full video/audio and the segment may be different

	var fizzbuzz = function(n){
	var answer = '';
	if (n % 3 === 0) {
	answer = 'Fizz';
	}
	if (n % 5 === 0) {
	answer += 'Buzz';
	}
	if (!answer){
	answer = n;

	var fibonacci = function(){
	var _cache = [];

	var fib = function(i){
	if (i < 0) {
	console.err('invalid input. i must be >= 0');
	return;
	}
	if (i < 2) {
	_cache[i] = i;

	if (!require(phaseR)){
	install.packages('phaseR')
	require(phaseR)
	}

	# a system of two non-linear autonomous ODE's
	ode = function (t, y, parameters) {
	x <- y[1]
	y <- y[2]
	dy <- numeric(2)

	// by default, it fetches pdf
	// change pdf to an extension you want
	var pdflinks = [];
	Array.prototype.map.call(document.querySelectorAll("a[href$=\".pdf\"]"), function(e, i) {
	if ((pdflinks \|\| []).indexOf(e.href) == -1) {
	pdflinks.push(e.href);
	}
	});
	console.log(pdflinks.join(" "));
	// copy the logged file names