Esther Villars esthervillars

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.

Connect to your EC2 instance
Install zsh : sudo apt-get update && sudo apt-get install zsh
Edit your passwd configuration file to tell which shell to use for user ubuntu : sudo vim /etc/passwd
Look for ubuntu user, and replace bin/bash by bin/zsh
Install OhMyZsh : sh -c "$(wget https://raw.githubusercontent.com/robbyrussell/oh-my-zsh/master/tools/install.sh -O -)"
Disconnect from your instance and reconnect it.

Build and install Google Test

Download the latest_ (1.7.0) from Google Code (Q: where is it going to live, once GCode shuts down?)

Then follow the primer_, but more to the point, the README (YMMV) Having installed CLion and cmake, this is how I built gtest:

brew install cmake
cd gtest-1.7.0

	#Do not include this file until your production certificate is working
	#This config Forcing ssl will prevent lets encrypt from verifying your domain
	#https://docs.aws.amazon.com/elasticbeanstalk/latest/dg/https-singleinstance.html
	Resources:
	sslSecurityGroupIngress:
	Type: AWS::EC2::SecurityGroupIngress
	Properties:
	GroupId: {"Fn::GetAtt" : ["AWSEBSecurityGroup", "GroupId"]}
	IpProtocol: tcp
	ToPort: 443

	# Dont forget to set the env variable "certdomain", and either fill in your email below or use an env variable for that too.
	# Also note that this config is using the LetsEncrypt staging server, remove the flag when ready!

	Resources:
	sslSecurityGroupIngress:
	Type: AWS::EC2::SecurityGroupIngress
	Properties:
	GroupId: {"Fn::GetAtt" : ["AWSEBSecurityGroup", "GroupId"]}
	IpProtocol: tcp
	ToPort: 443

	/*!
	* Petrify
	* Redirect yourself to a permanent version of a GitHub blob or tree.
	*
	* Say you're looking at `https://github.com/documentcloud/documentcloud/blob/master/public/javascripts/embed/search_embed.js#L315-L317`
	* and you want to link to it in an email or GitHub comment. You realize that,
	* when people follow the link years or even days later, it's likely the file
	* will have changed so much that those line references won't be the same, and
	* people won't know what you were talking about.
	*

	# Change YOUR_TOKEN to your prerender token
	# Change example.com (server_name) to your website url
	# Change /path/to/your/root to the correct value

	server {
	listen 80;
	server_name example.com;

	root /path/to/your/root;
	index index.html;

	require 'spec/support/grep_matcher'

	describe do
	disallow_presence_of pattern: "send(.*#",
	location: "app/",
	description: "Do not use dynamic method invocations",
	failure: "Please change dynamic method call to something more sane."
	end

	# add a new "put_copy" method to the Amazon client's S3Object class
	# to enable copying an object from 1 bucket to another
	# http://docs.aws.amazon.com/AmazonS3/latest/API/RESTObjectCOPY.html
	class AWS::S3::S3Object
	def self.put_copy(source_key, target_key, source_bucket, target_bucket, options = {})
	original = open(url_for(source_key, source_bucket))
	default_options = { :content_type => original.content_type }
	store(target_key, original, target_bucket, default_options.merge(options))
	acl(target_key, target_bucket, acl(source_key, source_bucket))
	end