The Semantic Precision for AI Retrieval of Knowledge (SPARK) PubMed/PMC Challenge rewards teams developing reproducible AI retrieval and evidence synthesis systems to ease researchers’ biomedical literature searches.
Funder: NIH Challenges
Due Dates: January 1, 2027 (Platform registration deadline) | January 15, 2027 (NIH registration and code upload deadline) | January 31, 2027 (Final submissions)
Funding Amounts: $250,000 total prize pool; each task awards $100,000 first place and $25,000 second place
Summary: Rewards reproducible AI systems for citation-backed biomedical literature retrieval and evidence synthesis using a designated PubMed/PMC corpus.
Key Information: Cash prizes require U.S. citizenship/permanent residency or an eligible U.S. entity; proprietary hosted LLM APIs are prohibited.
The National Library of Medicine (NLM) runs this prize competition, co-sponsored by NIH’s Chief AI Officer and Office of Data Science Strategy. It seeks AI-powered systems that move beyond keyword searches to understand scientific meaning, prioritize robust evidence, and reduce researchers’ time and cognitive burden when navigating biomedical literature.
Participants build an end-to-end system addressing one or both tasks. Task 1, Known-Item & Provenance Retrieval, retrieves specific documents satisfying natural-language information needs. Task 2, Open-Ended Exploratory Information Needs, synthesizes evidence across multiple sources into concise, objective, contextually accurate responses. Outputs must use valid PubMed/PMC citations and relevant supporting passages from the challenge corpus. Task 2 responses must explicitly address conflicting findings and scientific disagreements.