Content Recommendation Engine reference profile.
Which content or products should this user see next, ranked by expected engagement or conversion?
Overview
Increases engagement time and session frequency through more relevant recommendations.
Who owns this decision
VP Product, Head of Personalization
What this Specialist Model actually does
Decision: Which content or products should this user see next, ranked by expected engagement or conversion?
The final taxonomy records 6 supporting or external components, but it doesn't promote an Oumi-owned specialist model for this engine.
Who uses this Specialist Model
- VP Product
- Head of Personalization
- Chief Product Officer
- Director of Growth
Where the specialist model sits
The final taxonomy records 6 supporting or external components, but it doesn't promote an Oumi-owned specialist model for this engine.
Business-rules layer
Supporting component
Business-rules layer for inventory, licensing, and promoted-placement constraints
How Oumi connects to your stack →
Real-time feature store
Supporting component
Real-time feature store for session and freshness signals
How Oumi connects to your stack →
Two-tower or collaborative-filtering retrieval model
Supporting component
Two-tower or collaborative-filtering retrieval model for candidate generation
Learned ranking model (gradient-boosted trees or deep ranking network)
Supporting component
Learned ranking model (gradient-boosted trees or deep ranking network) for final scoring
Content embedding model
Supporting component
Content embedding model for cold-start and content-based similarity
How Oumi connects to your stack →
Specialist LLM
Supporting component
Specialist LLM for recommendation explanations and conversational recommendation queries
How to evaluate and improve it
Evaluation
- Click-through rate and conversion rate on recommended items
- Watch-time or session-length lift versus control
- Catalog coverage / long-tail exposure
Feedback signals
- Click-through and watch/purchase completion on recommended items
- Explicit feedback (thumbs down, "not interested," ratings)
- Session abandonment after a recommendation set is shown
- Long-term retention and return-visit correlation with recommendation quality
- A/B test outcomes comparing recommendation strategies


