…
…
…
…
NOT A SCREENSHOT. THIS IS THE RUNNING APP
…
…
…
NOT A SCREENSHOT. THIS IS THE RUNNING APP
Everything here is live software I designed, built, deployed and still keep running: 48 apps across products, daily tools and AI experiments.
Most of the intelligence in these apps runs on local models on my own GPUs. Ollama and llama.cpp handle text, embeddings and classification, ComfyUI does the image work, and Kokoro and a few other open-weights models cover speech. Cloud models are an option and never a dependency. Where they turn up it is usually as a paid upgrade or a bring-your-own-key lane, so the free path costs electricity.
Scores, prices, totals and verdicts are computed in code and never guessed by a model. The model explains, phrases and drafts. The code decides. It started in Humanize and spread through Repomate, Skills, Autoblog, Fin and Cryptobot.
Two GPUs and a small fleet of hardened containers: every app runs non-root on a read-only filesystem with dropped capabilities and its own resource limits, behind a reverse proxy that terminates TLS. One of these apps was compromised through a stale framework CVE in August, and that incident is why the whole portfolio is sandboxed this way now.
Every app is loaded live in the frame. Use the arrows or your keyboard to move between them, All 48 to jump straight to one, and Read more for the full story behind whichever app you are looking at.