Agentic News - Decode News from Different Perspectives

Break every world affair into a multi-agentic perspective, have the agents debate their views & form your opinion! Breaking the "Polarization" & "Narratives" for every world affair.

Comment

🌲 Inspiration

Today's news media(print or otherwise) are pushing their "narratives" & are harming the democracies of the world. In a healthy democracy, every voter has the right to know all the facts, presented in an unbiased way, & then only form an opinion. This is resulting in polarised societies where voters are easily persuaded with these high polarization news coming from multiple news sources.

In print media, or otherwise, news articles have, over the last 10 years, started to extensively report biased information, leading to gullible voters, & its only through X(Twitter) or Reddit discussions that we are able to get the different opinions. This is hurting our democracies.

In my country, India, we have debating sessions on Prime Time News channels, but though there are panelists coming from different perspectives and party lines, these debating sessions are moderated in such a way that not all the participants are given a fair chance to have their point. Thus leading to fragmented societies, where different perspectives are lost due to this "Noise".

⚙️ What it does

My app solves this issue by having 3 agents, one with a conservative view, one with a Progressive view & one with a Neutral view, where every news article, news link, YouTube video etc are independently analyzed by these 3 agents in parallel, & then they have a live debating session among themselves to present their view. The user acts like a Jury, & can ask any agent their view point for a topic being discussed. These agents independently identify commonalities and disagreements in a given news article and generate a Polarization Index to guide debate on these points.

Multi-Lens Analysis: Every article is processed by three distinct agents:

  • The Sentinel: Analyzes through a right-leaning, market-focused perspective.
  • The Advocate: Analyzes through a left-leaning, equity-focused perspective.
  • The Jurist: A neutral fact-checker that strips away adjectives and identifies missing context.

Prism Synthesis: A final "Orchestrator" agent cross-references all outputs to find the "Shared Reality" and calculates a Polarization Index.

In the current digital news media, we have an audio option to hear the news article, but that's just a plain narration. We made an app that goes a step further, & instead has the news article analyzed by these agents and then presents you with their views from different perspectives, & has a debate among themselves.

Thus, having the News link as the "white light", NewsPrism acts like a "prism" and breaks the article into different perspectives(aka colours), thereby fostering different perspectives to voters, to hear every perspective to form their opinion.

🧱 How we built it

  • Frontend: React 19, Tailwind CSS 4, Motion (Framer Motion), Lucide React.
  • Backend: Express.js (Node.js), Axios, Cheerio (for web scraping).
  • AI Engine: Amazon Bedrock featuring Amazon Nova 2 Pro (for complex reasoning) and Amazon Nova 2 Lite (for rapid-fire debate responses).
  • Infrastructure: Amazon Web Services (AWS).

The Infrastructure: AWS

To ensure NewsPrism is production-ready and scalable, we leveraged the AWS ecosystem: AWS App Runner: The entire application is containerized and deployed via App Runner. This allows us to scale automatically based on traffic while keeping latency low for our global users.

AWS Secrets Manager: We securely handle our Amazon Bedrock credentials and other sensitive environment variables, ensuring our integration is both powerful and secure.

Amazon Bedrock: Used to orchestrate the agentic workflow, utilizing the high-speed inference of the Amazon Nova 2 model family to maintain a fluid, real-time debate experience.

Architecture: https://github.com/StnkRB/Nova-NewsPrism/blob/main/ARCHITECTURE.md

🧗 Challenges we ran into

Developing the app within the constraints of initial model quotas was a challenge. We recognize the immense potential of this tool, but scaling it to serve voters globally requires robust infrastructure and funding. Transitioning our agentic workflows to the Amazon Nova architecture required significant prompt engineering. Since we were moving from other models, adapting to Nova's specific reasoning patterns was an exciting learning curve, but ultimately resulted in much faster response times for our live debate feature.

🏅 Accomplishments that we're proud of

We tested the agents with highly biased news links from recent global escalations and political topics. We were amazed at how well the Amazon Nova models maintained their grounded personas, presenting viewpoints solely based on the facts presented in the article (grounding) without hallucinating external biases. We are thrilled to see these agents work together to present a 360-degree view of any news article. Moving beyond plain narration to a dynamic, multi-perspective debate is a massive leap forward for news consumption.

📚 What we learned

Agents can be more consistent than humans in conveying a specific, grounded perspective. They aren't subject to the same external pressures or shifting biases that humans face over time. The beauty of using Amazon Nova is the ability to define a persona that remains steadfast and logical, providing a far more reliable "prism" for analysis than a human counterpart might.

🔭 What's next for NewsPrism

We want to realize and develop this app for a global audience. We have ideas to extend this agentic flow further—perhaps incorporating real-time social media sentiment analysis via AWS Lambda—to provide even deeper context to the "Shared Reality" our agents seek to find.

If you see the advantage that world democracies will gain from informed, critical-thinking voters, help us take NewsPrism to the next level!

Built With

Share this project:

Updates