It’s been a two-week ‘sprint’, so I thought I would share a quick update.
If you don’t know what/who Pagey is, check out my previous blog post about him here.
Basically, I built a personal book recommender (Štivoje) to get good book suggestions every week or so, based on my ‘taste profile’ built on ca. 320 books I’ve liked, and compared to 52.5k+ books from a dataset. He was built with Python, some JS, n8n, Google Sheets etc. and was a personal tool for myself.
Two weeks ago, I jotted down some potential next steps – such as making him smarter, prettier, pivoting (i.e. to music recommendations), or packaging into ‘micro-SaaS).
I decided to focus on ‘making smarter‘ and ‘packaging into micro-SaaS‘.
Additionally, in the meantime, Štivoje’s gone international, changing his name (and passport) to Pagey.
Main updates and learnings
- Used Lovable to build a landing page and launched PageyBooks beta
- Made an n8n flow I can easily recreate to accomodate my first users (ca. 20 signups so far, and it takes me ca. 3-5 minutes to ‘onboard’ new users). Onboarding:
- get user’s book list (and enrich it if needed e.g. with book descriptions)
- run the script for TF-IDF vectorization and cosine similarity using Google Colab
- sort results and take the top 1000 books
- duplicate n8n flow to use for the particular user
- Signed up for payment processor (Lemon Squeezy)
- Need to register products (Pagey packages) and start selling once everything is ready
- Need to register products (Pagey packages) and start selling once everything is ready
- Expanded the initial dataset with a cutoff at 2020 by writing a Python script to scrape the top books for each year from Goodreads (some 750-1250 books per year, 3000-5000 new titles)
- Did some discovery and got early user input:
- for instance, it’s tedious for people to have to write down or add their books into an Excel – I will make a ‘Netflix-like’ process as an option for users to just click books they’ve liked instead