KDD Cup CRAG 2024 · Silver Medal
Second-place solution for Meta's CRAG challenge, delivering reliable multi-hop retrieval and controlled generation.
Competition Repo →Flagship Projects
Systems, Benchmarks, and Knowledge Resources
Each project pushes Retrieval-Augmented Generation forward—from competition-grade systems to public benchmarks and surveys.
PruningRAG · CIKM 2025
Plug-and-play framework with multi-source pruning strategies that filter noisy knowledge before generation; accepted to CIKM 2025.
Paper & Framework →HoH Benchmark · ACL 2025
Measures how outdated information hurts RAG reliability and now part of the ACL 2025 Main Conference program.
ArXiv →Knowledge-Oriented Survey
Comprehensive taxonomy and evaluation review for RAG systems, covering mechanisms, applications, and open problems.
Survey Repo →Agent-R1 · 900★
End-to-end reinforcement learning agent that couples retrieval, planning, and execution for complex tasks; community support now exceeds 900 GitHub stars.
Framework →Timeline
Milestones in Retrieval-Augmented Generation
2024
Meta KDD Cup · CRAG Track
Multi-stage retrieval, reranking, and controllable decoding achieved a silver medal among 1,400+ teams.
2024
PruningRAG · CIKM 2025
Multi-granularity pruning policies for source selection, now an accepted paper at CIKM 2025, reducing hallucinations in deployed RAG agents.
2025
HoH Benchmark · ACL 2025
First dataset to stress outdated knowledge, now accepted by the ACL 2025 Main Conference, quantifying freshness effects on downstream reasoning accuracy.
2025
Knowledge-Oriented RAG Survey
Comprehensive taxonomy that catalogs retrieval strategies, evaluation lenses, and safety considerations for production RAG systems.
2025
Agent-R1 · 900★ Community
Open-source Agent-R1 crosses 900 stars as researchers adopt its retrieval-aware RL training loop for trustworthy autonomous agents.
Resources
Toolchains, Data, and Community
Competition Toolkit
Modular retrievers, rerankers, and decoders extracted from the KDD Cup pipeline for rapid experimentation.
Pruning Policies
Multi-source heuristics that cull redundant passages before feeding them into the generator.
Temporal Benchmarks
HoH-style datasets that capture changing facts and measure model freshness.
Awesome RAG Papers
Curated reading list and taxonomy for anyone building RAG systems in production.
Why Retrieval Still Matters
We prioritize grounded reasoning by constraining large language models with curated, verifiable context. Our work spans pruning, benchmark design, and agent training to keep answers safe and up-to-date.
Responsible Systems
Multi-source pruning and audits reduce hallucinations and encourage evidence-backed responses.
Continuous Evaluation
Automated freshness tests and safety audits keep RAG deployments aligned with evolving knowledge and risk requirements.
Open Collaboration
Repositories, dataset releases, and surveys lower the barrier for the broader RAG community.