He changes one variable at a time and watches what moves, and he has a good ear for ambiguity — he was the one who worked out that a machine and a person do not agree on whose “right” is meant. That kind of noticing is difficult to teach.
Book Nook
Keshav Nair · Lilly Robson · Kenzo Harris
The problem they picked
Most recommendations come from engines built to sell you the next thing, which is why they keep pointing at what you have already read. A reader trying to get out of their own rut — or to buy from the shop down the street instead of a warehouse — is largely on their own.
Book Nook recommends by conversation rather than catalog rank, and sends every suggestion to the libraries and independent bookstores nearby, alongside what other readers actually thought of it.
“We built this to expand literacy — wider access to diverse books, encouraging readers to explore new genres, and supporting lifelong learning. And because we connect readers to local bookstores and libraries instead of just selling them a book, we help small businesses get discovered too, not just Amazon.”
Book Nook, at the showcase
How it was built
Automation Workflows
Stella, the recommendation chat, runs on an n8n workflow that searches several book sources at once.
API Integration
Leaflet and OpenStreetMap for the nearby map, Photon for keyless geocoding, and Open Library as the catalog fallback.
Database Management
Reader memory — history, ratings, library, community shelf — through Supabase database functions.
n8nAutomation Platform
SupabaseBackend & Database
Leaflet + OpenStreetMapMapping
Open LibraryExternal API
Also in the buildConversational recommendation · Community shelf and ratings · Nearby library and bookstore search
