Personalised Discovery PoC
Make the front page learn from the viewer, not just the crowd.
Select interests, run a four-card learning round, then watch the 5 x 5 grid update using explicit choices, freshness, live/VOD preference, thumbnail hook quality, watch events, skipped items, and not-interested feedback.
Loading profile...
Step 1
Choose the first signal
These category choices initialise the recommendation string before watch history exists.
Step 2
Four-card learning round
Fresh candidates from chosen categories. Skips and clicks both teach the model.
Step 3
Recommended watching
25 organic cards scored by affinity, freshness, hook quality, engagement, novelty, and penalties.
Short-form surface
Shorts
Uses the same profile, but with faster feedback and higher novelty rotation.
Separated paid surface
Sponsored and paid-for content
Targeting can use the profile, but paid placement stays clearly labelled and isolated from organic ranking.
Manual browsing retained
Standard categories
Users can still browse outside the recommendation model.