LaDiariaRAG: Newsroom Agents for la diaria
A specialized dynamic retrieval and search agent system for the Uruguayan news portal la diaria — designed as an article-discovery platform that encourages reading and deep engagement, not another summarization bot.
The Problem
Newsrooms need discovery tools that surface related reporting, not shallow summaries that replace reading. Agentic workflows forget prior editorial context across turns (“agentic amnesia”). Heterogeneous archives require retrieval that stays faithful to published journalism.
Approach
Custom MCP servers to give the journalistic team persistent tools and context. Retrieval oriented toward article discovery and engagement loops. Architecture that treats editorial memory as a first-class concern alongside generation.
Technical Stack
- Custom MCP servers for persistent context
- RAG over heterogeneous news archives
- Agent orchestration for article discovery
- TypeScript / Node.js as primary runtime
Outcome
An AI layer that supports the newsroom’s craft — finding the right piece, keeping context, and pushing readers deeper into the archive — instead of collapsing coverage into disposable chat answers.