E-Commerce AI Sandbox
An e-commerce sandbox combining Vertex AI Search with a Gemini pipeline that enriches and feeds the product search index.
- Status
- prototype
- Year
- 2025
- Stack
- GCP Vertex AI, Vertex AI Search, Gemini
Why this project
Keyword search on a typical e-commerce catalogue breaks down as soon as a query is phrased in natural language or contains a typo. This sandbox tests how far Vertex AI Search, combined with a Gemini pipeline that enriches product sheets upstream, can bring internal search closer to an engine that understands intent rather than just text.
Use cases
- Natural-language search across a product catalogue
- Gemini-driven enrichment of product sheets before indexing (missing attributes, rewording, business synonyms)
- Similar-product suggestions by semantic proximity rather than manual tags
How it works
A Gemini pipeline processes each product sheet to enrich it (attributes, synonyms, rewording) before it feeds the Vertex AI Search (VAIS) index. User search then runs directly through VAIS, which benefits from this upstream enrichment instead of relying on the search engine alone.
What I took away
The quality of semantic search depends as much on upstream data preparation (the Gemini enrichment step) as on the search engine itself: a poorly fed index gives poor results, even with a capable engine like VAIS.