Dr. Christian Geuer-Pollmann chgeuer

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.

running:

bash create-vod-hls.sh beach.mkv

will produce:

    beach/
      |- playlist.m3u8
 |- 360p.m3u8

ffmpeg

Converting Audio into Different Formats / Sample Rates

Minimal example: transcode from MP3 to WMA:
ffmpeg -i input.mp3 output.wma

You can get the list of supported formats with:
ffmpeg -formats

Convert WAV to MP3, mix down to mono (use 1 audio channel), set bit rate to 64 kbps and sample rate to 22050 Hz:

Latency numbers every programmer should know

L1 cache reference ......................... 0.5 ns
Branch mispredict ............................ 5 ns
L2 cache reference ........................... 7 ns
Mutex lock/unlock ........................... 25 ns
Main memory reference ...................... 100 ns             
Compress 1K bytes with Zippy ............. 3,000 ns  =   3 µs
Send 2K bytes over 1 Gbps network ....... 20,000 ns  =  20 µs
SSD random read ........................ 150,000 ns  = 150 µs

Read 1 MB sequentially from memory ..... 250,000 ns = 250 µs

Elixir Native UI - Boyd Multerer https://www.youtube.com/watch?v=77FW-jrCyCs

Only dependencies are Elixir, Erlang, and glut to access OpenGL.
No UI threads! Individual processes can be used to track mouse
Slides track mouse as long as button is down https://youtu.be/77FW-jrCyCs?t=37m35s
Sliders can be rotated https://youtu.be/77FW-jrCyCs?t=40m4s
500 lines animating still a few ms to render https://youtu.be/77FW-jrCyCs?t=43m34s

Thinking In Ecto - Darin Wilson https://www.youtube.com/watch?v=YQxopjai0CU

Repository pattern https://youtu.be/YQxopjai0CU?t=2m12s
Explicitness https://youtu.be/YQxopjai0CU?t=5m29s

Backend Architectures

Twitter

ror, scala, jetty, erlang, thrift, mongrel, comet server, my-sql, memchached, varnish, kestrel(mq), starling, gizzard, cassandra, hadoop, vertica, munin, nagios, awstats


	import argparse
	import sys
	import grp
	import pwd
	import subprocess
	import os
	import json

	class PreRegistration:

	#!/usr/bin/python

	'''
	Python implementation of passcode hashing algorithm used on the Samsung Galaxy S4 GT-I9505 4.2.2
	Correct PIN for hash and salt below is 1234.

	Get 40-character hash value in ascii hex format from file /data/system/password.key on the phone

	Get salt in signed numeric format by doing sqlite3 query SELECT value FROM locksettings WHERE name = 'lockscreen.password_salt' on /data/system/locksettings.db

	using System;
	using System.Collections.Generic;
	using System.Reflection;
	using System.Threading.Tasks;
	using System.Web.Http;
	using System.Web.Http.Dispatcher;
	using System.Web.Http.SelfHost;

	namespace SelfHost
	{