20:00 / day
Concentrated Liquidity
One daily 20-minute window where everyone is online at once. Connection happens in a tight band, not a 24/7 scroll. Scarcity makes attention real.
// independent dev — palo alto, cartx 4070 online
AlejandroGonzalez
Independent AI Developer & Local LLM Hacker. I build privacy-first software and run the models on my own silicon.
Everything ships local-first. No clouds I don't control, no data I can't see, no inference I can't run at 2am on a single GPU.
01Core project
Sonder
A privacy-first, introvert-native social network.
Most social apps reward the loudest, most-online people. Sonder is built for the rest of us: small windows, real signals, and a network that goes quiet when you do. The data never leaves a place you control.
ai-assisted
Spark Matching
An on-device model aligns people by genuine interest and vibe, not engagement bait. Matches are suggested, never sold.
consent-first
No Cold DMs
Nobody slides into anything. Conversation only opens after a mutual spark, so the inbox stays a place you actually want to open.
Progression status04 / 04 · live
02Other projects
Things I'm shipping on the side
Suomi Academy
Language learning gameA gamified Finnish learning game that turns one of Europe's hardest languages into something you actually keep opening. Streaks, spaced repetition, and a story you only unlock by speaking.
- gamified
- spaced-repetition
- finnish
Tunee.net
Generative music video engineAn AI creative tool that turns a track into a full music video - generating visuals that move with the audio. Part of a small suite of creative tooling I build for musicians.
- generative
- audio-reactive
- creative-ai
03Stack & specs
Languages & frameworks
- 01RustTauri v2
- 02Reactfrontends
- 03Tailwind CSSstyling
- 04PythonML / glue
How it runs
Local LLM pipelines
Ollama and llama.cpp running quantized models on a single RTX 4070. Inference stays on the desk, not the cloud.
Custom MCP servers
Hand-rolled Model Context Protocol servers that wire local tools and data into the models I run.
Privacy-first by default
Tauri desktop builds, on-device inference, and a hard rule: if the user can't see where the data goes, it doesn't ship.
~/rig — neofetch
- host
- workstation / linux
- gpu
- nvidia rtx 4070 · 12gb
- runtime
- ollama + llama.cpp
- context
- custom mcp servers
- desktop
- tauri v2 (rust)
- status
- online
ag@rig:~$ ollama run llama3
04Contact
Building something local-first? Let's talk.
I take on a small number of collaborations around on-device AI, privacy-first products, and creative tooling. The fastest way to reach me is below.