Our current recommendation engine keeps surfacing out-of-stock items

Understood. That usually means the recommender is optimizing for relevance without a hard availability filter.

Quick fixes:

1. Filter out-of-stock items at serving time
   Apply a hard constraint before ranking or display.

2. Use real-time inventory in the feature set
   If stock changes frequently, ensure recommendations use the latest stock signal, not a stale batch snapshot.

3. Add stock status as a ranking feature or penalty
   Down-rank low-stock or unavailable items instead of only filtering.

4. Separate candidate generation from eligibility
   Generate broadly, then remove items that are unavailable before final ranking.

5. Monitor freshness and cache TTLs
   If inventory is cached too long, recommendations will drift out of sync.

6. Add guardrail metrics
   Track:

   • % of recommended items out of stock
   • click-through on in-stock vs out-of-stock
   • conversion loss due to stock issues

If you want, I can help you design a simple architecture for inventory-aware recommendations or draft the exact ranking/filtering logic.