Media Party Barcelona 2026 didn’t end when the main stage lights went down at CaixaForum. On September 9, the conference added a third day entirely dedicated to building: the Media Party Hackathon, hosted at BIT Habitat Barcelona in the Poblenou district and powered by Cloudflare as official sponsor.
What the hackathon is
Media Party has always organized its programming around three pillars — Discover, Learn, and Do — and the hackathon is where the “Do” happens. Rather than a closed-door coding marathon for programmers only, it was framed as an interdisciplinary lab bringing journalists, developers, designers, and strategists together around a single problem: how newsrooms adapt to an AI-driven, post-search information landscape.
The sprint focused teams on four themes: workflow automation, agentic journalism, synthetic-content verification, and information integrity. Cloudflare backed the day with technical grants and resources aimed at helping the strongest prototypes graduate into sustained, production-ready open-source tools.
The competing teams
Four projects made it to the final round of pitches:
- ESCUCHA — An interactive web application built around an AI voice agent that interviews and converses with users to understand their news consumption habits and preferences. The tool processes that dialogue to identify and categorize audience information needs, helping media outlets deliver more relevant, personalized content.
- Meta-Discovery API for Agents — Infrastructure designed to help AI agents discover and surface publisher content in search, aiming to increase trust in agent-mediated discovery and give publishers an incentive to make their platforms agent-readable.
- AIdas (AI Drift & Alignment Scanner) — A tool that detects bias across politically contested topics by comparing how different large language model providers answer the same questions, surfacing where and how model outputs drift from one another.
- FACTHELP — A search and filtering assistant built for fact-checkers and verification reporters. Instead of manually clicking through dozens of links to piece together what’s actually being said about a breaking story, FACTHELP narrows and prioritizes results, filtering out old, unrelated, or misleadingly similar stories so reporters spend less time chasing noise and more time verifying.
The winner
AIdas took the top prize, recognized for tackling one of the thorniest open problems in AI-mediated information: measuring and exposing ideological bias and inconsistency across LLMs on contested political topics — a capability increasingly relevant as more readers get their news filtered through AI systems rather than direct search.
Together, the four finalists captured the throughline of the day: newsroom tools built not by developers working alone, but by mixed teams solving problems that combine editorial judgment, product design, and technical execution.
