A multimodal retail search and recommendation application powered by Google Cloud Spanner, Vertex AI (Gemini), and Streamlit.
- Multimodal Search: Search by text, vector (semantic), or hybrid (text + vector).
- Native Spanner Enhancements: Uses Spanner's built-in
SEARCH(..., enhance_query=>true)for automatic query expansion (synonyms, spell correction). - Generative AI Integration:
- Gemini 1.5 Flash: Analyzing user queries and generating personalized responses.
- Vector Search: Using Vertex AI web-gecko text embeddings for semantic retrieval.
- Graph-Based Recommendations: leveraging Spanner Graph (SQL/PGQ) to recommend related products based on purchase history (e.g., "People who bought THIS also bought...").
- Resilient Architecture:
- Systemd service for automatic restarts.
- Robust error handling and "Nuclear Patch" for known client-side metric issues.
- Frontend: Streamlit
- Database: Google Cloud Spanner (Graph + Vector + Search)
- AI/ML: Vertex AI (Gemini 1.5 Flash, Text Embeddings)
- Infrastructure: Google Compute Engine (Debian 11)
See DEPLOY_INSTRUCTIONS.md for detailed deployment steps.
- Search: Use the sidebar to switch between "Hybrid", "Vector Only", "Full Text Search", etc.
- Debug: Enable the "Debug Panel" to view raw SQL queries, execution times, and graph traversals.
- Recommendations: Click on any product to see "People who bought this also bought..." recommendations driven by Spanner Graph.