Hugging Face has announced how its core services power the recently revived Papers with Code platform, a search engine designed to make open AI research more accessible and discoverable. The platform helps researchers find artifacts related to papers, locate state-of-the-art results across AI domains, and discover related work that can build on existing research. The search infrastructure combines keyword and vector-based search methods to understand complex queries that don't match exact paper titles or abstracts. The Papers with Code search system uses a hybrid retrieval architecture that combines PostgreSQL's full-text search capabilities with dense vector embeddings for semantic understanding. Hugging Face provides three core services supporting this infrastructure: Jobs for batch embedding of the paper corpus, Storage Buckets for durable artifact storage between systems, and Inference Endpoints for low-latency query embedding at request time. The system maintains embeddings for over 110,000 papers from arXiv and Daily Papers, using Qwen3-Embedding models pinned to specific versions with 256-dimensional vectors optimized for speed. The architecture separates expensive offline corpus-building work from the lightweight online search service, with automatic fallback to full-text retrieval if the embedding endpoint is unavailable. The implementation treats embedding formats as versioned APIs, recording exact model revisions, dimensions, normalization methods, and content hashes to prevent subtle vector mismatches. This design prioritizes both system resilience and search quality, addressing the unique challenge of finding research papers based on semantic meaning rather than exact keyword matches.