All work
AI · 2023
Pennaro Retail
Personalised recommendations and semantic product search.
+18% average order value
- Services
- AI & Data, Backend
- Timeline
- 5 months
- Team
- 4 people
- Year
- 2023
01The challenge
Keyword search returned poor results for natural queries, and every shopper saw the same generic recommendations.
02What we built
Semantic search built on embeddings with keyword fallback, plus a recommendation model trained on browsing and purchase history and tested with controlled experiments.
Stack
- Python
- OpenSearch
- PyTorch
03Outcomes
+18%
average order value
+24%
search-to-purchase rate
60 ms
p95 search latency
04Inside the build