How Hugging Face Inference Endpoints, Jobs, and Buckets Power Search on Papers with Code

A hybrid search system was built for the Papers with Code website to improve how users find AI research papers. The system combines traditional keyword search using PostgreSQL with semantic vector search using pgvector and the Qwen3 embedding model. It leverages Hugging Face Jobs for offline batch processing of the paper corpus, Storage Buckets for durable artifact storage, and Inference Endpoints for real-time query embedding. This architecture separates high-throughput offline tasks from low-latency online services to ensure a fast and robust search experience. The results from both keyword and vector search are combined using the reciprocal rank fusion algorithm to deliver more relevant results.