Jekyll2026-04-29T21:20:54+00:00https://kwoolford.github.io/feed.xmlKyle WoolfordPersonal site for funKyle WoolfordCreature Creator2026-04-25T00:00:00+00:002026-04-25T00:00:00+00:00https://kwoolford.github.io/project/creature-creatorCreature Creator (working title: Morphlings Arena) is a small Python 3D game prototype built with the Ursina engine. The project drops a set of randomly generated creatures — each assembled from primitive geometric shapes — into an enclosed arena where they wander, chase rivals, fight, and periodically morph their appearance and stats. The aim is a self-contained playable demo, intentionally limited in scope to keep iteration fast.

Purpose

Explore procedural creature design and lightweight 3D combat mechanics using pure Python, without external 3D assets or a heavy engine. Ursina sits on top of Panda3D and keeps the setup to a single uv add command.

Highlights

  • Creatures built entirely from Ursina primitives — spheres, cubes, cones — with randomized body parts (eyes, horns, tails, wings)
  • Per-creature randomized stats: color, size, speed, health, attack damage, aggression range
  • Periodic morphing every few seconds that visibly changes a creature’s scale, color, and stats mid-match
  • In-scene health bars; creatures are removed when health reaches zero
  • Player controls: Space spawn, R reset arena, P pause, Esc quit

Technical notes

Item Detail
Language Python 3.12+
Engine Ursina 8.3+ (Panda3D)
Package manager uv
Entry point uv run python main.py

Needs more detail in the README.

]]>Kyle WoolfordRecipe App2026-04-25T00:00:00+00:002026-04-25T00:00:00+00:00https://kwoolford.github.io/project/recipe-appRecipe App (Pantry Pairing Recipe App) is a single-file Streamlit application that generates personalized recipe suggestions using a locally running LLM via Ollama. Users describe what’s in their pantry, their time budget, meal type, effort level, dietary needs, and flavor mood — and the app prompts the model to return three structured recipe cards complete with ingredient breakdowns, step-by-step instructions, and pairing suggestions. All inference runs locally; no cloud API keys are required.

Purpose

Test whether a minimal local-LLM stack (Streamlit + Ollama) can deliver a genuinely useful everyday cooking tool, while keeping the prototype simple enough to run anywhere with a single command.

Highlights

  • Configurable Ollama model selection in the UI — swap gemma4, llama4, or any pulled model without touching code
  • Structured Markdown output per recipe: time estimate, pantry ingredient match, extra items needed, numbered steps, fun drink/side/vibe pairings
  • Optional favorites saved to a local JSON file
  • Graceful error message and instructions when Ollama is not running

Technical notes

Component Detail
Frontend Streamlit
LLM runtime Ollama (local HTTP localhost:11434)
Data validation Pydantic
HTTP client requests
Package manager uv
Entry point uv run streamlit run app.py
]]>Kyle WoolfordBeat Saber Automapper2026-04-18T00:00:00+00:002026-04-18T00:00:00+00:00https://kwoolford.github.io/project/beatsaber-automapperAn open-source AI system that takes an audio file and produces a playable Beat Saber level — notes, arcs, chains, bombs, obstacles, and a synchronized light show — packaged as a v3-format .zip. The pipeline is a three-stage transformer: a shared audio encoder feeds an onset detector, a note-sequence decoder, and a lighting decoder, each conditioned on difficulty, genre, and per-frame song-structure features.

Purpose

The goal is to replicate what good human mappers do — density planning, swing-direction flow, lighting that tracks song energy — from audio alone, with a model small enough to train overnight on a single RTX 5090. Existing automappers either target the obsolete v1 format, skip lighting entirely, or produce maps that feel mechanical. This project targets v3, includes lighting, and is trained on a curated high-rating slice of BeatSaver.

Highlights

  • Style-cohort training (V5). Replaced a single averaged model with per-mapper style cohorts — 18 mappers in 9 style buckets — each trained independently.
  • Auto-researcher harness. A YAML queue of training specs runs overnight; each result is scored on a playability + style-closeness composite and written to a leaderboard.
  • Rich conditioning. Every stage receives difficulty + genre + song-structure embeddings (RMS, onset strength, band energies, section id, section progress) so soft sections slow down and drops get dense.
  • Token-level note generation. Stage 2 uses an autoregressive transformer over a 183-token vocabulary, with beam search or nucleus sampling and an ergonomics loss that penalizes parity violations.

Technical notes

Stage Task Arch Loss
1 onset detection 6-block TCN + 2L transformer BCE on fuzzy labels
2 note sequence 8L transformer decoder CE + flow + ergo
3 lighting events 4L transformer decoder CE + label smoothing

All stages share one audio encoder: 4-layer CNN frontend → sinusoidal positional encoding → 6-layer transformer (d_model=512). Trained end-to-end on Beat Saber map data — no pretrained speech weights, because what matters here is low-level rhythmic structure, not semantics.

Note. Trained with a RTX 5090 (sm_120) requires PyTorch nightly with cu128; stable wheels don’t yet compile for Blackwell.

]]>Kyle WoolfordWelcome2026-04-18T00:00:00+00:002026-04-18T00:00:00+00:00https://kwoolford.github.io/note/welcomeThis site is a running log of the projects I’m working on and miscellaneous things I want to keep track of.

Most entries will be short digests auto-generated from project READMEs; some will be notes I drop in by hand. The homepage surfaces featured projects at the top, with the full archive filterable below. Each project page is meant to read like a notebook entry — a single-sentence summary, a stack row, a status, then sections for purpose, highlights, and technical notes.

If something is missing or broken, the source is on GitHub.

]]>Kyle Woolford