Truth Field Detector

Real-time narrative risk assistant that detects emotional manipulation and misinformation patterns in news, video, and social media using Amazon Nova.

Comment

Truth Field Detector: Defending Human Perspective in the Age of Narrative Manipulation by Eloisa Flores :-)

Inspiration In an era where digital narratives are increasingly optimized to bypass human critical thinking, I felt a deep responsibility to empower the individual. Manipulation is often subtle, hiding behind emotional triggers and absolutist claims. I was inspired to create a tool that restores Human Agency — not by judging what is 'true,' but by illuminating the mechanisms of influence. My goal is to turn the 'digital noise' back into a clear field of truth, ensuring that technology serves as a shield for the mind rather than a tool for steering it.

What it does Truth Field Detector analyzes news articles and text in real time. It detects Emotional Language, Absolutist Claims, and Missing Citations — the three pillars of narrative risk. It delivers a Narrative Risk Score (0–100) through a multilingual analyzer, providing instant visual feedback and natural voice summaries.

## How we built it This is a fully functional proof-of-concept powered by Amazon Bedrock. - Analysis: I used Amazon Nova 2 Lite for high-speed, real-time narrative reasoning. -Audio: Natural voice feedback generated via Amazon Polly. -Infrastructure: The Backend was built on Flask, hosted on Render, with auto-deploy from GitHub. -Frontend: A clean, responsive UI designed for immediate user clarity. I used Canva to design the UI/UX (User Interface), then implemented it using Python and CSS.

Challenges we ran into I initially explored Nova 2 Sonic for voice, but the bidirectional WebSocket streaming was incompatible with my architecture — so I pivoted to Amazon Polly, which provided more stable voice feedback. I also implemented a text-input fallback to bypass scraping blocks on certain news URLs, and resolved Git branch conflicts between local and remote.

Accomplishments that we're proud of Fourteen days ago, I had never touched Python. As an Independent AI Safety Researcher, I went from fearing 'Skynet' to building two functional AI safety tools in just two weeks — one for Amazon Nova and one for Gemini. I taught myself the full stack because the mission of Truth and Connection couldn't wait for a manual. I don't just research alignment; I build the tools to enforce it. Creating a multilingual analyzer that is live and accessible to anyone in the world is my first step in proving that qualitative research can be translated into scalable code. I am particularly proud of maintaining the focus and rigor to complete this project while navigating extreme real-world constraints and domestic challenges — proving that the mission of AI Safety can thrive outside the vacuum of a quiet lab.

## What we learned I learned how Amazon Bedrock manages structured AI responses, the technical foundations of narrative risk metrics, and how to deploy a production-ready AI application from scratch. Most importantly, I learned that "High Agency" is the most powerful tool in any developer's stack.

What's next for Truth Field Detector My roadmap includes a Browser Extension for real-time browsing protection, mobile integration, and expanding the analysis to video and audio streams to combat multimodal misinformation.

Built With

Share this project:

Updates