Jongmin Park GzuPark

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.

The code for this tutorial is here

Opencv provides are useful, but limited, method of building a GUI. A much more complete system could be acheived using pyqt. The question is, how do we display images. There are quite a few possible routes but perhaps the easiest is to use QLabel since it has a setPixmap function. Below is some code that creates two labels. It then creates a grey pixmap and displays it one of the labels. code: staticLabel1.py

from PyQt5.QtWidgets import QWidget, QApplication, QLabel, QVBoxLayout
from PyQt5.QtGui import QPixmap, QColor
import sys

	{
	"art_style_profile": {
	"style_name": "Modern Isometric 3D Icons – High-Fidelity",
	"visual_elements": {
	"shape_language": "Real objects reduced to smooth, softly-rounded forms with gentle bevels, yet retaining all signature sub-shapes (e.g., pasta strands individually twisted, grill marks precisely scored).",
	"colors": {
	"palette_strategy": {
	"base_hue": "Use the object’s most iconic colour.",
	"tone_triplet": "Three tones of that hue (Lightness ±18 %).",
	"pop_accent": "One high-saturation complementary / triadic hue for fun highlights.",

	# Taken from: https://gist.github.com/sayakpaul/23862a2e7f5ab73dfdcc513751289bea

	from diffusers import FluxPipeline, FluxTransformer2DModel
	from transformers import T5EncoderModel
	import torch
	import gc


	def flush():
	gc.collect()

	import logging
	import os
	import sys
	import traceback
	from contextlib import contextmanager

	import diart.operators as dops
	import numpy as np
	import rich
	import rx.operators as ops

	ARG UBUNTU_VERSION=20.04
	ARG CUDA_VERSION=11.3.1
	ARG CUDA=11.3
	ARG CUDNN_VERSION=8

	FROM nvidia/cuda:${CUDA_VERSION}-cudnn${CUDNN_VERSION}-devel-ubuntu${UBUNTU_VERSION}
	LABEL maintainer "http://gzupark.dev"

	ARG CUDA_VERSION
	ARG CUDA

	import os
	from urllib.parse import urlparse
	import urllib.request
	import zipfile
	import tempfile
	from subprocess import call


	zip_url = 'https://github.com/konlpy/konlpy/archive/refs/heads/master.zip'
	dirname = 'konlpy-master'

	from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
	import tensorflow as tf

	model_name = 'bert-base-cased'
	tokenizer = AutoTokenizer.from_pretrained(model_name)
	model = TFAutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2)

	texts = ["I'm a positive example!", "I'm a negative example!"]
	labels = [1, 0]

	'''
	A simple Time Timer app made of ShapeNodes.
	<Change Line 120 for seconds or minutes counting>
	<Uncomment Line 126 and fill in your shortcut name to
	automatically run other shortcut you defined
	after timer stopped>
	Odd touch attemps will start the timer
	Even touch attemps will stop the timer
	Stops automatically when the set times passed
	'''

	"""Hack to add per-session state to Streamlit.

	Usage
	-----

	>>> import SessionState
	>>>
	>>> session_state = SessionState.get(user_name='', favorite_color='black')
	>>> session_state.user_name
	''