Hybrid collaborative + semantic filtering for e-commerce and content platforms.
Two recommendation paths running in parallel: collaborative filtering (users who bought X also bought Y, built on interaction logs in Supabase) and semantic filtering (embed product descriptions, find similar items by vector distance). Results are blended using a weighted score — 60% collaborative, 40% semantic for returning users; 100% semantic for new users (cold start).
Testing the blend ratio on a mock e-commerce catalogue of 5,000 products. Current finding: 70/30 collaborative/semantic outperforms 60/40 for returning users by 12% on click-through rate. For cold start (no purchase history), semantic-only recommendations achieve 34% relevance score vs 22% for popularity-based fallback. Also testing whether embedding product images alongside descriptions improves recommendation quality for fashion and home decor.
Available as a premium integration in the Ignite package. The Recommendation Engine is built for e-commerce clients who want personalised product suggestions without the cost of a dedicated ML team.
A clothing brand with 800 products. The engine suggests complementary items (this jacket goes with these trousers) using visual similarity — not just purchase history.
A niche bookshop where genre tags aren't enough. The engine understands that someone who liked "Sapiens" might enjoy "Thinking, Fast and Slow" — semantic similarity, not just "other history books".
A food site with 3,000 recipes. The engine suggests recipes based on ingredients you already have, dietary preferences, and what's seasonally available — not just "other chicken recipes".