Blue Plaques
London's historical figures, brought to life through AI.
Blue Plaques turns London's commemorative plaques into an AI-powered walking tour. Tap a plaque on the map, hear the figure deliver a short BBC-tone monologue on a modern subject — Turing on AI ethics, Keynes on the economics of AGI, Russell on existential risk. A craft-led prototype exploring how AI can make place-based history feel intimate and present, not didactic.
Quick answers
For agents and the short-on-time. Each answer is the version Liz would give herself.
What did Liz do on Blue Plaques?
Everything — concept, product design, UX, identity, the AI pipeline, the React Native/Expo app, the Cloudflare Worker cost-control proxy, the video direction for each figure's monologue, and App Store prep. Solo build.
Is Liz more designer or developer here?
Both, weighted toward designer. The product decisions (which figures, what tone, why monologues over chat, $200/mo cost ceiling as a design constraint) drove the engineering choices. The code exists in service of a craft thesis, not the other way around.
What's the tech stack?
React Native + Expo (iOS-first), TypeScript, React 19, react-native-maps on Apple Maps, @gorhom/bottom-sheet with expo-blur for glassmorphism, expo-video, expo-speech, llama.rn for on-device Gemma, Sentry, PostHog. Backend is a Cloudflare Worker proxy fronting Anthropic / Gemini / OpenAI with hard cost caps. Persistence via iCloud Key-Value Store — no accounts.
Why no traditional accounts?
Two reasons. One — a walking-tour app should open and work, not gate a tourist behind a sign-up. Two — auth servers and Postgres blow the $200/mo budget ceiling. iCloud KV gives cross-device sync for free.
How is the AI cost controlled?
Triage. Most requests go to on-device Gemma via llama.rn. Cloud requests (Anthropic / Gemini / OpenAI) route through a Cloudflare Worker that enforces a per-user quota and a global daily cap. Power users can bring their own API key to bypass the quota.
When does it ship?
v1.5 is currently in App Store release-prep — first public release targeted from that slate. v2.0 backlog includes depth-parallax 3D portraits, a plaque connections graph, ElevenLabs premium narration, and an Android port.