Sep 2026
Getting this site up
Wanted somewhere to put what I'm working on that wasn't just a Discord message
or a tweet that disappears. Still figuring out what goes here exactly.
I build AI companions, VR tools, local LLM setups, and whatever else won't leave my head. This is where I share the work — and the rest of it too.
VRChat tools, Python apps, self-hosted infra — what I'm shipping.
My local AI VTuber — LLM, voice, Twitch, all running on my own hardware.
Diary, thoughts, dreams, childhood, condition — the honest stuff.
Things I build — mostly because nothing else did what I needed. Some are finished, most are always changing.
Fully local AI persona streaming on Twitch as buhbwah. Tsundere anime girl character inspired by Neuro-sama. Local LLM via Ollama, custom voice through GPT-SoVITS, speech input via faster-whisper. Tuned to react to what's actually happening in the game rather than hallucinating game state.
tkinter control panel tying together MicBoost, STT, and PTT — all outputting over OSC into VRChat. How I manage my entire voice setup in VR without touching the keyboard.
Real-time phone-to-PC camera streaming over HTTPS and WebSockets. Turning my phone into a proper wireless webcam without any third-party apps.
Electron app using MediaPipe pose landmarking to drive VRChat avatars over OSC — camera-based FBT, no dedicated trackers needed.
Flask app that serves a file exactly once then deletes it. Tunneled through Cloudflare — private sharing that self-destructs after the first view.
Python SteamVR overlay for dimming the Meta Quest 2 display from inside VR using the controllers. Small QoL thing I use constantly.
Owncast behind a Cloudflare Tunnel with OBS as source. Private stream — no Twitch, no platform, just a URL I share with who I want.
The biggest challenge was keeping her reactive — early versions hallucinated hero names and ability descriptions that were obviously wrong to anyone watching. The prompt now keeps her relational and reactive to what I report, rather than generative about things she can't know.
The voice stack was the hardest part. GPT-SoVITS had cascading dependency issues — torchcodec DLLs, a Gradio/Jinja2 conflict, and the FFmpeg build all needed sorting before the API ran stable. Took a while, but it sounds good now.
The goals, real ones and vague ones. Things I want to build, learn, figure out.
Ideas that stuck around long enough to write down — about tech, building stuff, whatever.
Honest writing about where I'm at — physically, mentally, whatever's going on.
Things I remember, how I grew up, the stuff that shaped how I think.