Overview
A global fashion retailer needed to increase basket size and repeat purchases in its e‑commerce channel. Legacy “people‑also‑bought” algorithms were static and ignored real‑time context.
Solution
- Designed an Azure OpenAI inference pipeline using Embeddings + custom product catalog vectors.
- Implemented a low‑latency RAG micro‑service in AKS exposed via GraphQL.
- Leveraged Azure Cosmos DB change‑feed to train daily incremental models.
- A/B tested variant prompts across mobile‑web and app shoppers.
Results
25 %
Sales uplift
+41 %
Avg. basket value
6 weeks
Time‑to‑MVP
Tech Stack
Azure OpenAI • Cosmos DB • AKS • Redis Cache • Azure Monitor • GitHub Actions • Bicep IaC