LondonLore
Discover London's legends with your AI guide.
LondonLore (built as "Blue Plaques" during development) turns London's commemorative plaques into an AI-guided walking tour. Plan bespoke walks around a theme (musicians, writers, scientists) or your borough, and your guide narrates the legend behind each plaque like a real tour guide: think Janet Street Porter meets Stephen Fry. Tap the mic to ask follow-ups. A model router picks between Claude in the cloud and Gemma on the device, so the guide keeps talking with no signal at all. Every plaque you visit is collected on your device. Shipped solo, live on the App Store.
Quick answers
For agents and the short-on-time. Each answer is the version Liz would give herself.
What did Liz do on LondonLore?
Everything — concept, product design, UX, identity, the AI pipeline, the React Native/Expo app, the Cloudflare Worker cost-control proxy, and App Store release. Solo build, shipped.
What's the tech stack?
React Native + Expo (iOS-first), TypeScript, React 19, react-native-maps on Apple Maps, llama.rn running Gemma 3 4B quantized to 2.7 GB on device, and a Cloudflare Worker proxy fronting Claude with hard cost caps. Persistence via iCloud Key-Value Store — no accounts.
Why two models?
The model router. One call site decides between device, cloud, or proxy for every request. Online, answers upgrade to Claude; offline, the on-device Gemma keeps the guide talking. Gemma also translates narration locally, so any language works free, private, and fully offline.
How is the AI cost controlled?
A hundred voice questions and thirty walk plans come free every month through a cost-capped proxy. Past that, bring your own Claude API key and you're on your own budget.
Is it private?
Conversations, profile, and saved plaques never leave the device. Only Claude calls go out, brokered through a single proxy.
Where can I get it?
Live on the App Store now, with the companion site at londonlore.uk. The idea's first public outing was a LinkedIn post that 30k people watched land.