Booksmart 2.0
Architected and built four modules across Markono Group’s storefront and its print factory, under IMDA’s Digital Leaders Programme.
The problem
Markono Group, a Singapore printer and distributor, sells more than five million titles through its Booksmart storefront. Nobody browses five million books by category, so a reader who knew what they wanted to read about, but not which book, had no real way in. Behind the store, supplier feeds changed shape without warning, and the print floor was scheduled from a spreadsheet and one person’s experience.
The solution
I architected and built four modules, under IMDA’s Digital Leaders Programme:
- Search that finds a book by title, author, keyword or ISBN. Describe a subject instead, and it hands back a reading list.
- Sasha, a reading-concierge chatbot. Tell her what you are in the mood for and she suggests books, only ever from the catalogue.
- An AWS Glue transformation layer for the supplier feeds. It runs on a schedule, so nothing sits idle between runs, and when a feed changes shape the job fails where someone can see it, rather than letting bad data slip downstream.
- Data-driven print scheduling in Markono 4.0, fed by live floor status instead of the spreadsheet. IMDA credits Markono 4.0 as a whole with an immediate 30% boost in manufacturing productivity.
Sasha only recommends what the store stocks
Sasha is Booksmart’s chatbot: a reader describes what they are after and she suggests books. I treated it as a retrieval problem before a chat one. She has to find real titles in the catalogue first, and only then talk about them. The failure I cared most about was a fluent, convincing answer about a book the store does not carry, so every recommendation comes from the catalogue.