Hi, I'm Sky: privacy tools, tiny computers, and neural nets in CSS

Hi everyone, I'm Sky. I'm a solo developer and the founder of Quantara, a one-person lab that builds public-good tools around harm reduction and privacy and publishes skeptical security and AI analysis. Quantara operates under Slateweir Works LLC.

A few things I've made:

• narcan.delivery: a state-by-state guide to naloxone access across all 50 US states, with an open dataset behind it.
• qwennie and diffusion.css: a 78k-parameter chat transformer and a small diffusion model that both run entirely in CSS, no JavaScript.
• unlisted and metadetect: local-first privacy tools. One reduces location clues in photos taken at home, and the other detects nearby recording devices such as smart glasses over Bluetooth.
• ESP32 firmware: a Doom-style raycaster on a 0.42-inch OLED, an H.264 benchmark on the ESP32-P4, and an audio-reactive glitch camera.

I also train and fine-tune small models, mostly ones that fit on hardware people already own. Everything I ship is free and openly licensed.

I'm here to trade notes on small models, on-device inference, and privacy tooling, and to see what everyone else is building. Happy to answer questions about any of the above.

Everything lives at metrix187.github.io.