The challenge
Translate loosely expressed shopping needs into useful results without allowing an AI model to invent products or specifications.
The approach
Extract a structured search intent, then apply deterministic catalogue filters and ranking. Embeddings support semantic retrieval, while catalogue-grounded Q&A helps answer questions about available listings.
Engineering focus
Explicit sort instructions, numeric constraints and fallback behaviour matter as much as the model itself. This project explores where flexible language interpretation should meet predictable application logic.
Project context
Built with Next.js App Router, TypeScript and Tailwind CSS for an FDE assessment. Detailed source-level diagrams will be added after reviewing the final repository.