A Four-Layer Multi-Agent Architecture for Automated Journalism: Event-Driven Orchestration with Hybrid Context Management

Darshika pundir, Tuhin Sharma, Arun Chauhan

Keywords: Multi-agent systems, Listener-aware communication, LLM orches- tration, Event-driven architecture, Context orchestration, Scalable AI systems, Retrieval-augmented generation, Automated journalism

TL;DR: A four-layer architecture enabling specialized AI agents to collaboratively produce verified news articles through event-driven coordination and hybrid context management.

Abstract:

We present a four-layer architecture for automated journal-
ism addressing unbounded context growth, agent coordination complex-
ity, and quality assurance at scale. Our system features Layer 0 (Ob-
servability), Layer 1 (Specialized Agents), Layer 2 (Event-Driven Or-
chestration with Listener-Aware Communication), and Layer 3 (Hybrid
Context Management with RAG), enabling specialized agents to collab-
oratively produce publication-ready articles. The architecture employs
asynchronous message queues, agent pooling, and a context layer combin-
ing structured compression with sliding windows for agent interactions
and vector database retrieval (RAG) for unlimited source material han-
dling. This hybrid approach combines semantic compression (avg 56%
reduction, empirically validated on MCP responses) and RAG-based re-
trieval for unbounded source handling, while maintaining bounded agent
memory through progressive compression. Empirical validation on 100
news stories demonstrates 100% success rate with avg 36 second process-
ing time. The architecture enables concurrent story processing through
agent pooling and event-driven orchestration while maintaining quality
through multi-step verification and revision capabilities

Submission Number: 3