LondonLore
Discover London's legends with your AI guide.
The thesis
Place-based history should feel intimate and present, not didactic. London’s 837 blue plaques are quiet markers most people walk past. LondonLore gives them a voice: a guide who knows every legend, never runs out of patience, and keeps talking in the middle of the Underground.
From post to product
It started as a LinkedIn post that hit 30,000 views with one question: what if AI could bring history to life? The early prototype explored AI-video monologues from figures like Turing, Keynes, and Russell. The shipped product kept the soul of that idea and traded the video for something better: a live guide who narrates any of the 837 plaques in character and answers whatever you ask next.
What’s shipped
- Bespoke themed walks (musicians, writers, scientists, or your borough), planned by the AI guide
- In-character narration for every plaque, with slide-up cards for the full story and sources
- Voice follow-ups: tap the mic and ask anything
- Full offline mode via on-device Gemma 3 4B — no signal needed
- Automatic plaque collection: your personal LondonLore history, stored on device
- Free tier (100 questions and 30 walk plans a month), then bring your own key
The constraint that shaped everything
A hard monthly ceiling on AI costs. The architecture follows directly: on-device Gemma via llama.rn for offline and fallback, a Cloudflare Worker proxy in front of Claude for cost-capped routing, bring-your-own-key flows for power users, and iCloud Key-Value Store for sync with no accounts, no auth server, no Postgres. The cost ceiling isn’t a footnote, it’s the spec.