RAGTUM at SemEval-2026 Task 8: Contextual Query Rewriting and Dense Retrieval for Multi-Turn RAG

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

Summary

The RAGTUM system, presented at SemEval-2026 Task 8, addresses the inherent challenges of multi-turn Retrieval-Augmented Generation (RAG) by combining context-aware query rewriting with a dense retrieval strategy. Developed by a team from the TUM practical course, this system implements a robust pipeline. It begins by cleansing noisy corpora to ensure data quality. Subsequently, it utilizes dense OpenAI embeddings, managed through Milvus, for highly robust information retrieval. For both standalone query generation and the final response synthesis, RAGTUM employs models from the Gemini 2.5 flash family. This integrated approach effectively demonstrates the power of combining high-precision retrieval with fact-based generation across diverse domains, as detailed in pages 1784–1790 of the workshop proceedings.

Key takeaway

For Machine Learning Engineers building multi-turn RAG systems, this work highlights a practical architecture to improve conversational AI. You should consider implementing context-aware query rewriting to maintain conversational coherence and integrate dense retrieval using tools like Milvus with OpenAI embeddings for robust information access. Furthermore, exploring the Gemini 2.5 flash family of models for both query generation and final response synthesis could significantly enhance your system's fact-based generation capabilities across varied domains.

Key insights

The RAGTUM system enhances multi-turn RAG via context-aware query rewriting and dense retrieval with specific model and embedding choices.

Principles

• Context-aware query rewriting improves multi-turn RAG.
• Dense retrieval enhances RAG robustness.
• Corpora cleansing is crucial for RAG quality.

Method

The RAGTUM pipeline cleanses noisy corpora, uses dense OpenAI embeddings via Milvus for retrieval, and employs Gemini 2.5 flash models for query generation and response synthesis.

In practice

• Integrate query rewriting for conversational RAG.
• Use Milvus with dense embeddings for retrieval.
• Consider Gemini 2.5 flash for RAG synthesis.

Topics

• Multi-Turn RAG
• Contextual Query Rewriting
• Dense Retrieval
• OpenAI Embeddings
• Milvus Vector Database
• Gemini 2.5 Flash

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

Related on AIssential

• Benchmarking Retrieval Strategies for Biomedical Retrieval-Augmented Generation: A Controlled Empirical Study — Cross-Encoder Reranking significantly improves biomedical RAG performance by enhancing contextual precision.
• RCEM: Robust Conversational Search EMbedder in Distributional Shift — RCEM distills LLM query rewriting into an embedder for robust, low-latency conversational search without explicit rewriting.
• The retrieval rebuild: Why hybrid retrieval intent tripled as enterprise RAG programs hit the scale wall — Enterprise RAG is shifting from simple vector search to complex hybrid retrieval for production agentic AI.
• An Efficient Approach for Answering Not Readily Attainable Questions for RAG-based Applications — This paper proposes an efficient approach for RAG systems to answer questions not readily found in retrieved data.
• Beyond Semantic Search: The Complete Guide to Advanced RAG with Milvus | Sagar Patil — Advanced RAG requires multi-stage retrieval, combining hybrid search with techniques like query rewriting and reranking for production reliability.
• RAG Is Not Dead. You’re Just Building It Wrong. — Optimizing RAG for production requires systematic engineering across all layers, prioritizing retrieval over LLM changes.

Open in AIssential →

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