AI platform engineering / portfolio case study

Lumen

An AI-native thin-client computing platform. The phone is a dumb window; every app lives on a home server; and an embedded AI build engine writes, hot-loads, and pushes new apps to the device — sometimes in the middle of the conversation that asked for them.

33live apps served from one hub
1thin WebView shell on the device
0app-store releases needed to ship an app
5machines behind the glass: GPU, edge, media, memory, agents
The thesis

The device is the wrong place for software to live.

A phone should be a sensor package and a pane of glass. Compute, state, memory, and intelligence belong on machines you control — where apps can be created, reshaped, and retired as fast as your life changes.

Design principle
Ship one shell, once. After that, "installing an app" is just the server learning a new trick — and every screen you own gets it instantly.
Evidence and boundaries

Operational evidence

Lumen runs daily as the author's primary phone environment: a ~2,000-line FastAPI hub, 33 plugin apps, 11 iterated APK builds of the shell, GPU speech/vision pipelines, WebAuthn fingerprint login, and a documented device-management protocol for thin-client laptops.

Portfolio-safe presentation

This case study uses representative descriptions and omits real hostnames, addresses, credentials, and private infrastructure details. Concept apps in the Vision section are clearly labeled as designs, not shipped software.