Today's progress on Conjure didn't involve a flashy UI feature. It was
infrastructure the recommendation algorithm that decides what content
users see when they open the app.
And I kept it intentionally simple.
The Algorithm
During onboarding, users select interest categories (Science, Development,
Design, etc.). Those interests are stored on their profile and become the
single source of truth for their feed.
When the homepage loads:
- 70% of the feed = posts matching the user's selected interests
- 30% of the feed = exploration content from other categories
No ML. No behavioural tracking. Just a binary category match against a
stored preference list, sorted by recency within each bucket.
Why So Simple?
Two reasons I'm not ashamed of:
1. Complex algorithms need data.
A sophisticated system needs enough user behaviour to produce meaningful
signals. I don't have that yet. A complex algorithm on thin data just
produces confidently wrong recommendations.
2. Premature complexity kills momentum.
If I tried to build something clever at day 6, I'd spend two weeks
architecting it, second-guessing edge cases, and shipping nothing.
Simple → ship → learn → improve is the only sane order of operations
at this stage.
What I Also Shipped Today
The homepage UI now renders a proper feed on login. No search required,
no empty state staring at you. Open Conjure, start reading immediately.
What Comes Next
The 70/30 split is a starting ratio I'll tune it once I have real usage
data. A learning mechanism that factors in reading history is on the roadmap,
but it earns its complexity only after the simple version gets validated by
real users.
If you've been thinking about writing on Conjure, now's a good time.
The more content exists, the better the recommendations get.
Top comments (0)