AMU at RAG4Reports 2026 Task B: A Practical Multilingual RAG Pipeline for Citation-Grounded Reports

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, short

Summary

AMU's submission to RAG4Reports 2026 Task B presents a practical multilingual retrieval-augmented generation (RAG) pipeline designed for evidence-supported report generation. This system integrates several components: full-query retrieval, optional query rewriting, dense retrieval utilizing Qdrant, cross-encoder reranking, diversity-aware context selection, and structured generation. The top-performing configuration employed BAAI/bge-m3 embeddings, BAAI/bge-reranker-v2-m3 for reranking, and gpt-5.1 for generation, using a medium reasoning effort and a partial-coverage prompt strategy. On the official leaderboard, the pipeline achieved an F1 score of 0.4351, sentence_support of 0.8280, and nugget_coverage of 0.3403, indicating strong grounding but limited comprehensiveness in its generated reports.

Key takeaway

For NLP engineers developing multilingual RAG systems for report generation, consider this pipeline's architecture to enhance evidence grounding. Your implementation could benefit from integrating BAAI/bge-m3 embeddings and BAAI/bge-reranker-v2-m3 for retrieval and reranking, paired with gpt-5.1 for generation. Focus on optimizing context selection and prompt strategies to improve report comprehensiveness, as indicated by the nugget_coverage score of 0.3403.

Key insights

A practical RAG pipeline combines diverse techniques for multilingual, citation-grounded report generation.

Method

The pipeline involves full-query retrieval, optional query rewriting, dense retrieval with Qdrant, cross-encoder reranking, diversity-aware context selection, and structured generation.

In practice

• Integrate BAAI/bge-m3 embeddings for dense retrieval.
• Utilize BAAI/bge-reranker-v2-m3 for effective reranking.
• Employ gpt-5.1 with partial-coverage prompting for generation.

Topics

• Retrieval-Augmented Generation
• Multilingual NLP
• Report Generation
• Qdrant
• BAAI/bge-m3
• gpt-5.1
• Cross-encoder Reranking

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Retrieval-Augmented Generation as a Bridge Between Classical NLP Pipelines and Modern LLM Systems — Retrieval-augmented generation (RAG) addresses the limitation of static knowledge in large language models (LLMs) by connecting them to external, up-to-date knowledge sources.
• Top 5 Embedding Models for Your RAG Pipeline — Effective RAG pipelines rely on embedding models that balance retrieval performance, real-world adoption, and deployment practicality.
• Grounded in Law: A Multi-Stage Anti-Hallucination Pipeline for Legal RAG Systems in Brazilian Portuguese — A multi-stage RAG pipeline significantly reduces legal citation hallucinations in Brazilian Portuguese LLMs.
• Issue #115 - Reranking in your RAG pipeline — Reranking enhances RAG pipeline precision by reordering initially retrieved documents using a Cross-Encoder.
• X-MADAM-RAG: Diagnosing and Handling Chinese-English Evidence Conflict in Retrieval-Augmented Generation — Multilingual RAG systems face significant challenges with contradictory evidence, requiring specialized diagnostic tools and pipelines like X-MADAM-RAG.
• ProvenAI: Provenance-Native Traces of Evidence in Generated Answers — Meaningful transparency in retrieval-grounded QA requires independently measuring correctness, citation fidelity, and source influence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.