Continuum

Every time a crisis hits the news— a school shooting, a climate disaster, a policy vote— What happens? People scroll, they get angry, and then…nothing. That’s not because people don’t care.

It’s because modern media has amnesia.

Headlines sell. News treats every event like a snapshot, which strips away the history that gives it meaning. The result is issue fatigue, recency bias, and outrage with no path to action. Google and Apple News prioritize clicks over context. Wikipedia gives you history, but doesn’t keep you in the loop.

Enter Continuum. We bring memory to the media.

Continuum is the personal assistant that guides you through current events. It equips you with the tools you need to stay informed, engage in meaningful discourse, and contribute to actual change. When a user is first onboarded, they can enter which issues they want to subscribe to. Our suggestion engine leverages Pinecone, k-means clustering, and LLM-based article classification to pinpoint trending topics. Each topic becomes a ledger—a persistent, living database.

After onboarding, the user is taken to the dashboard. The left panel contains each ledger the user is subscribed to. The center panel contains the timeline, which was generated by the ledger’s native web scraper. It utilizes semantic search and simple pre-processing to fill the user in on the events that have taken place within the issue. Behind the scenes after onboarding, the intelligent source selector has supplied the web scraper with a tailored list of targets for its corresponding topic in order to decrease search latency.

In the right panel of the dashboard, the RAG-enabled reasoning agent is put into action. It is a debate partner with the primary goal of helping you critically think about the issues at hand and come to your own conclusions. It uses LangGraph to maintain state across interactions, track user preferences, and generate arguments that synthesize information across time. To generate the summary in the top right, we utilized Keywords.ai's AI Gateway and Prompt Engineering platform to create a robust, hallucination-free model.

The driving force of Continuum is the context loop. At regular intervals, each ledger’s web scraper will update the reasoning agent’s context by adding pertinent information to the timeline. The AI’s analyses—and in turn yours—will shift and grow in near-real time.

Our most significant hurdle was defining and engineering our workflow to integrate AI. Each step was more complex than a simple LLM, requiring us to utilize modular design and ensure effective data flow between components.