Inspiration
In today's media landscape, news articles often deliver more emotion than information. We were inspired by the need for transparent, bias-free reporting that empowers readers rather than sways them. We set out to build "Neutral Ground" to help people access the objective facts quickly in real-time.
What it does
NeutralGround is an AI-powered tool designed to remove editorial bias and emotional language from news articles. It extracts factual points, cross-references them across multiple sources, and presents an unbiased summary. This helps users focus on the truth, without influence from political bias or persuasive wording. In addition, NeutralGround provides a list of the biased language and framing techniques it removed. This allows users to see exactly how the original content was shaped and what was taken out.
How we built it
For the frontend we used React and TypeScript. For the backend we used Flask and Express to process inputted articles and handle API requests. When a user submits a URL, the article is parsed with natural language tools and sent to Meta’s Llama 3 model through the Groq API. A prompt-engineered pipeline then extracts objective facts and filters out bias. The system is modular by design, making it easy to add features like multi-source validation and deeper contextual analysis in the future.
Challenges we ran into
There were a few challenges we ran into: -Handling CORS issues between the frontend and Flask backend -Prompt tuning to accurately remove bias without losing important factual content -Managing long article lengths with token limitations in free-tier LLMs -Dealing with variability in scraped article formats and quality
Accomplishments that we're proud of
-Successfully integrated an LLM pipeline that extracts and reformats biased content -Built an intuitive UI that makes critical media literacy tools accessible to everyone. -Designed a flexible backend architecture that can scale to include additional sources and verification logic -Using an open-source LLM model that allows managing long article lengths
What we learned
-Prompt engineering is a powerful tool. It takes iteration, precision, and contextual nuance to guide large language models effectively. -Gaining hands-on experience integrating AI models into a full-stack application, including communicating securely with LLM APIs like LLaMA via Groq. -Even subtle editorial choices in phrasing or structure can significantly shape how facts are perceived. -Creating ethical AI isn’t just about using the right tools but about making intentional decisions around how that technology is applied and the impact it has on end users.
What's next for Neutral Ground
-Expanding our bias-detection model to highlight specific framing techniques in real time -Building a browser extension to integrate fact-checking into live reading experiences -Automating cross-source reference and validation to strengthen fact integrity -Partnering with educational institutions and fact-checking organizations to promote media literacy at scale
Log in or sign up for Devpost to join the conversation.