alper alimoglu avatar-lavventura

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.

Writing tech articles

This is a description of how I write tech articles for various blogs. Hopefully someone else will find this useful as well.

Create a Gist for the article

When I begin writing a new article, I create a new [GitHub Gist][gist] for the article files. The Gist contains a file for the article text and code examples related to the article.

	;; help links
	;; https://github.com/magit/magit/discussions/4674#discussioncomment-2668452
	;; magit-blob-previous
	;; magit-blob-next
	;; * some links:
	;; __ http://www.mattduck.com/2021-05-upgrading-to-emacs-28.html
	(require 'package) ;; https://stackoverflow.com/a/14838150/2402577
	(setq message-log-max t)
	;; (when (eq system-type 'darwin)
	;; (setq debug-on-error t))

	uname_out="$(uname -s)"

	if [[ -r "${XDG_CACHE_HOME:-$HOME/.cache}/p10k-instant-prompt-${(%):-%n}.zsh" ]]; then
	source "${XDG_CACHE_HOME:-$HOME/.cache}/p10k-instant-prompt-${(%):-%n}.zsh"
	fi

	export COLORTERM=truecolor
	export DEFAULT_USER="$(whoami)"
	export ZSH=$HOME"/.oh-my-zsh"
	CASE_SENSITIVE="true" # Uncomment the following line to use case-sensitive completion.

	#
	# HackerRank - Cutting Metal Surplus
	#
	# The owner of a metal rod factory has a surplus of rods of arbitrary lengths. A local contractor offers to buy any of the factory's
	# surplus as long as all the rods have the same exact integer length, referred to as saleLength. The factory owner can increase the
	# number of sellable rods by cutting each rod zero or more times, but each cut has a cost denoted by costPerCut. After all cuts have
	# been made, any leftover rods having a length other than saleLength must be discarded for no profit. The factory owner's total profit
	# for the sale is calculated as:
	#
	# totalProfit = totalUniformRods × saleLength × salePrice − totalCuts × costPerCut

	#!/usr/bin/env python
	import argparse


	def main(command_line=None):
	parser = argparse.ArgumentParser('Blame Praise app')
	parser.add_argument(
	'--debug',
	action='store_true',
	help='Print debug info'

	;; Caveat: I'm an Elisp noob. I expect this will break if invoked before flycheck mode is possible, and am not sure that's possible.

	;; Invoke flycheck preferentially when flycheck-mode is enabled.
	;; My own workaround for https://github.com/commercialhaskell/intero/issues/268.

	(defun flycheck-or-norm-next-error (&optional n reset)
	(interactive "P")
	(if flycheck-mode
	(flycheck-next-error n reset)
	(next-error n reset)))

-;;; auto-remove.el --- Auto remove unused functions in python
-;;; Commentary:
-;; Uses external tool autoflake to remove unused imports from a Python file.
-;;; Code:
-(defcustom python-autoflake-path (executable-find "autoflake")
-  "Autoflake executable path.

	pb-kill-line () {
	zle kill-line
	echo -n $CUTBUFFER \| pbcopy
	}

	pb-kill-whole-line () {
	zle kill-whole-line
	echo -n $CUTBUFFER \| pbcopy
	}