LLM-Based Multi-Agent Blackboard System for Information Discovery in Data Science

Source: arXiv Authors: Alireza Salemi, Mihir Parmar, Palash Goyal, Yiwen Song, Jinsung Yoon, Hamed Zamani, Tomas Pfister, Hamid Palangi Date: 2025-09-30 Primary category: cs.MA All categories: cs.MA, cs.AI, cs.CL, cs.IR, cs.LG

Abstract

This paper revisits blackboard coordination for LLM agents in a data-discovery setting. Instead of a controller that must know every worker’s capability, a request is posted to a shared board and specialized agents volunteer responses based on their local competence and visibility. Across KramaBench plus modified DSBench and DA-Code variants, the authors report sizable end-to-end gains over stronger centralized baselines, which makes the blackboard pattern look like a live coordination option rather than historical AI nostalgia.