What programming/technical projects have you been working on?

I'm still working on my "repeatTest" property-testing framework. No new features, but I managed to unify two algorithms for tree search and delete a lot of code. I'm a lot happier with how it reads and more confident that it's correct.

I put together a little script to help run LLMs over multiple files.

It's all offline. No API keys needed. So far I've used it to suggest names for PDFs with poor names:

library llm-map  --rename --text ./gemma2.llamafile ~/Downloads/*.pdf

cat llm_map_renames.csv
Path,Output
/home/xk/Downloads/booka.pdf,/home/xk/Downloads/Mining_Massive_Datasets.pdf
...

This model works well but it is 20GB and takes about 20s per file:

wget https://huggingface.co/jartine/gemma-2-27b-it-llamafile/resolve/main/gemma-2-27b-it.Q6_K.llamafile
chmod +x gemma-2-27b-it.Q6_K.llamafile
./gemma-2-27b-it.Q6_K.llamafile

Gemini Nano might be another interesting one to play with

I'm curious what other use-cases people could have with something like this? The only other thing I can think of right now I can only think of text summarization

Currently trying to start up a blog. I’ve owned a domain for about 3-5 years now and never had anything on it, and I figure a blog is a good enough place to start. Especially with the IndieWeb resurgence.

Planning on using 11ty I think, though I’m a bit unsure what the best way to get started with templates is/if there are benefits to using one of the templating languages. (I followed a basic tutorial that just uses the .11ty.js for generating the html).

Its been a while since I posted, but I've spent a bit of time working on some terrain generation code, which has turned into a Minecraft mod that I'm tinkering with.

It's almost good to use, but a few remaining things like handing certain generation edge cases need to be dealt with, and adding a little UI data while the terrain is generating.

So help me God I will understand azure functions for F#

Haven't been doing much coding lately, but I recently decided that I should make a plugin to auto-repay large loans in OpenRCT2, since I like watching big numbers go down.

So I did just that.

The code is mostly quick and awful, but it does what I need it to. I'm happy enough with the result, especially considering that I haven't done Typescript stuff in forever.

Wrote a post about a small thing in MediaWiki software that people mess up all the time because no one knows how it works (but it's actually pretty powerful and very nicely designed)

Long story short, I'm writing a program for Linux that will let you script any input device with lua.

Long story long, you pass one or more devices evdev devices to the program. The program grabs the devices, and it will forward all events along with the device info for each event to a callback you implement in the lua config. It creates a virtual keyboard, mouse, and gamepad (I will perhaps explore other devices in the future and/or the ability for users to define custom virtual devices from their config) and from the lua script you can send events to these virtual devices which other applications will recieve the same as they would a physical device. So at the simplest, you can use it to remap devices such as remapping a gamepad to keyboard inputs for example, or you could use it to do things like consolidate multiple physical devices into a single virtual device. The ceiling for what you could do is limited only by your imagination (and willingness to actually implement it), layers, modes, context-sensitive inputs (e.g. "if you input X with your browser focused, perform action Y"), and whatever else you might want.

The initial idea was just for remapping gamepads with more complex features because I get frustrated by the limitations of even the more advanced systems for remapping like Steam's input system. Then I realized that all input devices go through the same kernel interface (there's even an evdev device for my power button so I may use that as a silly demo when I publish the project) so supporting other devices was not only easy but completely effortless.

Multiple devices at a time was just a random idea I had while working on it and I was initially not going to do it but I realized it would be make the program potentially much better as an accessibility tool, although I'm sure the overlap between "benefits from additional accessibility tools for games and such" and "wants to program their input device behavior with lua scripts" is probably pretty small, but I still hope it can be useful for people in that regard.

I was thinking that I need a way to quickly build spotify playlists. I asked chatGPT 4.x for some help.

• I provide tracks.txt -- the first line is the name of the playlist and the rest is artist - title
• based on the playlist name, it'll either create a playlist or update an existing one
• Its using the API with spotipy which has oAuth.
• it doesn't add duplicates
• It shows the progress. Not the prettiest thing, but it serves its purpose.

Summary Report:
Total tracks processed: 61
Tracks added: 49
Tracks skipped (already in playlist): 5
Tracks not found: 7
Tracks not found:
  - Les Baxter - Quiet Village Bossa Nova
  - Monster Rally - Coral
  - Lee Fields & The Expressions - Money Is King
Playlist 'Dinner Jazz by chatGPT' updated with 49 new tracks.

So that all works like a charm. I can do about 75 - 100 tracks without any issue. For the tracks themselves, I'm also asking chatGPT

Can you do the same for various genres of jazz including lounge, exotica, space-age pop, etc. Include artists like Martin Denny, Les Baxter, but also new bands like Khruangbin and Monster Rally. You can also include soul from Colemine Records like Aaron Frazer, Lee Fields, Durand Jones.

In this case it spits out the bands I referenced, so I ask it to branch out. The selection is pretty great.

It isn't perfect with apostrophes all the time, but for the most part its great.

import spotipy
from spotipy.oauth2 import SpotifyOAuth
import time
import sys

# Spotify API credentials
client_id = 'xxx'
client_secret = 'xxx'
redirect_uri = 'http://localhost:8888/callback'

# Authenticate with Spotify
scope = 'playlist-modify-public'
sp = spotipy.Spotify(auth_manager=SpotifyOAuth(client_id=client_id,
                                               client_secret=client_secret,
                                               redirect_uri=redirect_uri,
                                               scope=scope))

# Read tracks from the file
with open('tracks.txt', 'r') as f:
    lines = f.readlines()
    playlist_name = lines[0].strip()
    tracks = [line.strip() for line in lines[1:]]

# Function to search for a track and return its URI
def search_track(sp, artist, title):
    query = f'artist:{artist} track:{title}'
    results = sp.search(q=query, type='track', limit=1)
    tracks = results['tracks']['items']
    if tracks:
        return tracks[0]['uri']
    else:
        return None

# Check if the playlist already exists
user_id = sp.current_user()['id']
playlists = sp.current_user_playlists(limit=50)['items']
existing_playlist = None

for playlist in playlists:
    if playlist['name'] == playlist_name:
        existing_playlist = playlist
        break

# Use the existing playlist or create a new one
if existing_playlist:
    playlist_id = existing_playlist['id']
    print(f"Updating existing playlist: '{playlist_name}'")
    # Get the existing track URIs in the playlist
    existing_tracks = sp.playlist_tracks(playlist_id, fields='items(track(uri))')['items']
    existing_track_uris = {item['track']['uri'] for item in existing_tracks}
else:
    playlist = sp.user_playlist_create(user=user_id, name=playlist_name, description="")
    playlist_id = playlist['id']
    print(f"Creating new playlist: '{playlist_name}'")
    existing_track_uris = set()

# Add tracks to the playlist with rate limiting and progress display
track_uris_to_add = []
skipped_tracks = 0
not_found_tracks = []

for i, track in enumerate(tracks):
    try:
        artist, title = track.split(' - ')
        uri = search_track(sp, artist, title)
        if uri:
            if uri not in existing_track_uris:
                track_uris_to_add.append(uri)
            else:
                skipped_tracks += 1
                print(f"Track already in playlist: {artist} - {title}")
        else:
            not_found_tracks.append(track)
    except ValueError:
        print(f"Skipping malformed track entry: {track}")
        not_found_tracks.append(track)

    time.sleep(1)  # Rate limit to avoid hitting Spotify's API rate limits
    sys.stdout.write(f'\rProgress: {i+1}/{len(tracks)} tracks processed')
    sys.stdout.flush()

if track_uris_to_add:
    sp.playlist_add_items(playlist_id=playlist_id, items=track_uris_to_add)
    print('\nAll new tracks added to the playlist.')
else:
    print('\nNo new tracks were added to the playlist.')

# Summary report
print(f"\nSummary Report:")
print(f"Total tracks processed: {len(tracks)}")
print(f"Tracks added: {len(track_uris_to_add)}")
print(f"Tracks skipped (already in playlist): {skipped_tracks}")
print(f"Tracks not found: {len(not_found_tracks)}")
if not_found_tracks:
    print("Tracks not found:")
    for track in not_found_tracks:
        print(f"  - {track}")

print(f"Playlist '{playlist_name}' updated with {len(track_uris_to_add)} new tracks.")

Ultimately, I found over the last year or so that Spotify's recommendations are a bit of an echo chamber and I've been trying to pump new music into the algo... with little success. This has a lot of the stuff I already listen to, but I think attacking it via genre instead of specific artists will yield fresh music.

To quickly clear tracks.txt I use head -n 1 tracks.txt > temp.txt && mv temp.txt tracks.txt