---
title: "ReSUM: Reasoning-Driven Multi-Agent... | SI Paper Desk"
url: https://thesimind.si/papers/2611.09349/
description: "ReSUM splits meeting summarization into planned steps handled by separate SI agents, and the authors say this gives more faithful summaries than one-pass..."
---

Number 7 of 12 · November 9, 2026

# ReSUM: Reasoning-Driven Multi-Agent Framework for Query-Focused Summarization

ReSUM splits meeting summarization into planned steps handled by separate SI agents, and the authors say this gives more faithful summaries than one-pass methods.

Sonali, Deeksha Varshney and Asif Ekbal · Submitted November 6, 2026 · arXiv 2611.09349v1 · cs.MA

[Read the paper on arXiv](https://arxiv.org/abs/2611.09349) · [The PDF](https://arxiv.org/pdf/2611.09349) · [The day's 12](https://thesimind.si/papers/day/2026-11-09/)

## What they found

The authors propose ReSUM, a system where several language models work together. A planner directs one set of agents that find relevant parts of a meeting and another set that write the summary. The authors train the retrieval and summary agents by distilling reasoning from a teacher model. Distilling means teaching a model to copy the way a stronger model reasons. The abstract says this makes SI systems more robust on long inputs, cuts irrelevant searching, and gives more faithful and complete summaries. It says experiments show consistent gains on query-focused meeting summarization benchmarks, but it gives no numbers or named benchmarks.

## Why it matters

Query-focused summarization means answering a specific question about a long document, such as a meeting transcript. Meetings are hard because speakers interrupt, transcripts are long, and key facts are sparse. The abstract notes that even advanced models can omit information or make things up. Splitting the job into retrieval, evidence and writing steps is one way to cut those errors.

## How it changes the race

This is an incremental, task-specific result. It supports the idea that planned, modular SI agents can beat single-pass methods on long inputs, but the abstract gives too little detail to judge how large the gain is.

## Safety and control

The paper targets hallucination and information omission in summaries, which bear on how far people can trust SI output. The abstract gives no detail on how faithfulness was measured.

### The abstract, by the authors

LLMs excel at language tasks, but long-meeting QFS remains difficult due to complex speaker dynamics, sparse key information, and limited data. Meetings involve disfluencies, long transcripts and complex discourse. QFS have high risk of information omission, hallucination, even for advanced models such as GPT-4-mini and Gemini. Existing end-to-end methods compress retrieval, evidence extraction, and summary generation into a single pass, failing to handle the complex reasoning required for targeted meeting summaries. We propose ReSUM, a modular multi-LLM framework coordinated by a planner that orchestrates retrieval and summarization agents. By distilling teacher reasoning into query- and plan-conditioned retrievers and summarizers, ReSUM enhances long-context robustness, minimizes irrelevant exploration, and generates more faithful and comprehensive query-focused summaries. Experiments show consistent gains on query-focused meeting summarization benchmarks.

Quoted from arXiv:2611.09349. The research and its words belong to its authors.

Desk score **81** out of 100

Why it was picked

- Agents (in the title)
- Reasoning (in the title)
- Evals
- Multi-agent (in the title)
- Studies frontier models (GPT-4)
- Language models

Topics

- [Multi-agent systems](https://thesimind.si/papers/?tag=multi-agent)
- [Language models](https://thesimind.si/papers/?tag=language-models)
- [Retrieval](https://thesimind.si/papers/?tag=retrieval)
- [Reasoning](https://thesimind.si/papers/?tag=reasoning)
- [Training](https://thesimind.si/papers/?tag=training)

**arXiv ID:** 2611.09349v1
**Submitted:** November 6, 2026
**Fields:** cs.MA (Multiagent Systems)
**Level:** Intermediate
**Comments:** 10 pages, 4 figures

Fingerprint, SHA-256

`9ac97125e5bf36ff434f4f1261e7a83c5234422c4753de4a2256b4f4fa5f7556`

On Solana: [Proof of Mind seal, 437gF7...BR92](https://solscan.io/tx/437gF725DAG8gRKPBLFDmKiCEc8recBwc9f2nJbDSHirzNkcCnPsgSDCZzxXS15qHXCpomLuKrJuXUTtZbKhBR92)

### The exact text fingerprinted

{"arxiv":"2611.09349","version":1,"title":"ReSUM: Reasoning-Driven Multi-Agent Framework for Query-Focused Summarization","authors":["Sonali","Deeksha Varshney","Asif Ekbal"],"day":"2026-11-09","rank":7,"score":81,"status":"written","summary":{"takeaway":"ReSUM splits meeting summarization into planned steps handled by separate SI agents, and the authors say this gives more faithful summaries than one-pass methods.","found":"The authors propose ReSUM, a system where several language models work together. A planner directs one set of agents that find relevant parts of a meeting and another set that write the summary. The authors train the retrieval and summary agents by distilling reasoning from a teacher model. Distilling means teaching a model to copy the way a stronger model reasons. The abstract says this makes SI systems more robust on long inputs, cuts irrelevant searching, and gives more faithful and complete summaries. It says experiments show consistent gains on query-focused meeting summarization benchmarks, but it gives no numbers or named benchmarks.","matters":"Query-focused summarization means answering a specific question about a long document, such as a meeting transcript. Meetings are hard because speakers interrupt, transcripts are long, and key facts are sparse. The abstract notes that even advanced models can omit information or make things up. Splitting the job into retrieval, evidence and writing steps is one way to cut those errors.","race":"This is an incremental, task-specific result. It supports the idea that planned, modular SI agents can beat single-pass methods on long inputs, but the abstract gives too little detail to judge how large the gain is.","safety":"The paper targets hallucination and information omission in summaries, which bear on how far people can trust SI output. The abstract gives no detail on how faithfulness was measured.","level":"intermediate","tags":["multi-agent","language-models","retrieval","reasoning","training"]}}

[Check it on Verify a thought](https://thesimind.si/proof-of-mind/)

## The authors

Sonali, Deeksha Varshney, Asif Ekbal

Credit for this research belongs to its authors. SI MIND's page is a plain-English reading of their abstract on arXiv, not a review of the full paper.

Related on the desk

## More papers on these topics

[Nov 9, 2026 · Number 6Alignment by Confident Debate: Preference Learning from Multi-Agent DebateAbCD turns debates among several SI models into training preferences without gold answer labels, and the authors report better accuracy than ensembling, debate alone, or SFT on the same traces.AlignmentReasoningMulti-agent systems](https://thesimind.si/papers/2611.09568/) · [Nov 9, 2026 · Paper of the DayPhantom Consensus: Measuring Social Reasoning in Multi-Agent LLM Decision-MakingWhen SI models see ambiguous public signals from other models, they can drift to a group answer that no one privately held, and checking private judgments can catch this.Multi-agent systemsEvalsBenchmarks](https://thesimind.si/papers/2611.08971/) · [Nov 9, 2026 · Number 3AIM: Adaptive Interaction in Multi-Agent Debate for Multimodal LLM InferenceAIM lets SI agents in a debate choose, case by case, how to share evidence (text, graph, or image, audio and video regions), which raised accuracy and cut token use in the authors' tests.Multi-agent systemsMultimodalReasoning](https://thesimind.si/papers/2611.08950/)

[Number 6](https://thesimind.si/papers/2611.09568/) · [All 12 for Nov 9, 2026](https://thesimind.si/papers/day/2026-11-09/) · [Number 8](https://thesimind.si/papers/2611.09433/)