Maeil Business Newspaper × Anthropic AI Hackathon

Turning news into understanding

SkyDive designed and built a service to help readers understand the context and relationships behind news. Beyond reading individual articles, our aim was to show how events unfold and how people, companies, and other stakeholders relate to them.

Defining the problem and the experience

News is consumed quickly, but understanding its connections to earlier events and different stakeholder perspectives takes more work. Starting from Maeil Business Newspaper’s ‘Make Knowledge’ direction, we explored how fragmented information could become a reader’s own knowledge.

We first explored interpreting news through the thinking styles of well-known figures. We then focused on event history and stakeholder relationships to make the experience more accessible.

The concept extracted people, companies, and events from articles and connected their relationships. Buttons and overlays extended an existing news service, with core information available without signing in and analysis results reusable across readers. We also considered chat and public discussion as extensions.

Validating the design with data

We analyzed average article lengths to estimate Claude API processing costs and examined responses by category to prioritize development. Testing with real data showed that parts of the original pipeline did not fit, so we revised the core flow based on the results.

We also considered variation in AI stakeholder extraction. We focused on extraction criteria and output consistency so that the results would remain understandable across different kinds of news.

Implementation and verification

We used Claude Code during implementation, managed context, and recorded work in pull requests and commits. Tests and a separate testing interface helped us check functionality and whether data was reflected correctly.

Faster development made review more important. Unexpectedly lost functionality and harder-to-maintain data structures showed why AI-generated changes needed verification. We also prepared a product plan and promoted the service through Instagram Reels.

Results and lessons

We reached the finals and presented the result. Although we did not win a final award, we experienced the full product cycle: idea selection, requirements, data analysis, implementation, verification, and presentation.

We paid close attention to practical integration, operating costs, maintenance, and consistent AI output. Communicating the core value to a non-specialist audience within a short presentation was a separate challenge.

Behind the scenes

Seungmin Yang’s nine-part Korean-language Journal series documents the journey from analyzing the announcement and narrowing ideas to data analysis, implementation, and the final event. It is available in the Korean Articles section.