RAG APPLICATION
AI Shopping Assistant
Semantic product discovery powered by embeddings and vector search
A full-stack AI shopping assistant that uses semantic search, embeddings, and retrieval to help users discover relevant products through natural-language queries. The prototype combines a FastAPI backend, React frontend, ChromaDB vector storage, and OpenAI embeddings.
THE PROBLEM
Traditional ecommerce search often depends on exact keywords, category navigation, and rigid filters.
Users may know what they want conceptually without knowing the exact product name or dataset category, making natural-language discovery difficult.
THE SOLUTION
- Normalize product data into consistent, searchable documents.
- Create product and query embeddings with OpenAI's text-embedding-3-small model.
- Store and retrieve semantically similar product documents with persistent ChromaDB vector storage.
- Return structured results through FastAPI for display in a React conversational interface.
SYSTEM ARCHITECTURE
3 Components
- DummyJSON product data
- Field normalization
- Searchable document construction
SYSTEM VISUALS
WORKFLOW FLOW
IMPLEMENTATION STEPS
- 1.Ingest product data sourced from DummyJSON and normalize relevant attributes.
- 2.Combine titles, brands, categories, prices, ratings, availability, policies, descriptions, and reviews into searchable product documents.
- 3.Generate document embeddings with text-embedding-3-small and persist them in ChromaDB.
- 4.Embed each natural-language query with the same model and run vector similarity search.
- 5.Return relevant structured products through FastAPI.
- 6.Render retrieval results through the React chat experience and responsive product cards.
IMPACT & RESULTS
- Demonstrates the complete path from product data and embeddings to vector retrieval, API responses, and frontend results.
- Shows how semantic similarity can complement exact keywords in a product discovery experience.
- Connects a persistent vector database to a usable full-stack RAG-style prototype.