LLM Powered Product Search and Relevance Optimization to $9000

Bond Studio makes software that lets users capture their space with their phone and see a visualization of how that room could look if it were remodeled.  Users can browse products, select them, and see them visualized in their space.  

Part of this experience is search where users can type in queries and see products based on their query.  Currently this functionality is powered by a custom Python service that contains an index of all of our products and leverages LLMs to provide results to users.
Search quality and speed are critical to the overall user experience.
We have a comprehensive benchmark for this service and are looking to work with a software engineer to improve its performance on this benchmark.

This is not a greenfield project. You will be working with an existing production system and are expected to diagnose bottlenecks, propose improvements, and implement measurable gains.

You will analyze and improve a real-world product search system that includes:

A large, structured product catalog with many attributes and variants.
User-driven filtering and refinement (dimensions, style, finish, price, etc.).

Natural-language queries interpreted by an LLM-backed layer.
 

A backend search service that must be fast, stable, and scalable.
 

The focus of this engagement is on result quality while maintaining acceptable performance.