Tasks
Build and improve search systems under fixed resource constraints. Browse each task's goal, inputs, submission interface, and evaluation criteria.
All tasks
10 tasks- Reasoning-intensive Query Rewriting Build retrieval over a biology corpus by rewriting natural-language queries into precise search intent. query understanding
- Memory-constrained Dense Retrieval Search 1.5M document vectors while balancing retrieval quality, memory, index size, and latency. large-scale dense search
- Scientific-Paper Question Answering Answer scientific questions over 300 research-paper PDFs and identify the supporting paper. scientific RAG
- Long-PDF Evidence Localization Locate the physical pages in a single long PDF that best support a natural-language query. in-document search
- Long-document Reranking Improve a fixed reranker when relevant evidence lies beyond the first token window of a document. reranking
- Code Embedding Fine-tuning Improve a fixed-backbone code-retrieval embedder using the supplied training data and constraints. representation learning
- Query-side Encoder Alignment Improve query representations with a compact encoder while document vectors stay fixed and a runtime gate applies. query encoder optimization
- Agentic Search Optimize a ReAct retrieval starter to return five documents for multi-constraint BrowseComp-Plus queries. agentic search
- Sparse Retrieval Pruning Optimize candidate generation and postings pruning for opaque, unweighted sparse retrieval while preserving ranking quality. sparse postings · pruning · latency
- Rust BM25 Search Build a Rust BM25 engine over 8.8M passages and balance retrieval quality, query time, and index size against a fixed Pyserini reference. BM25 · Rust · indexing