About TruthLens
What inspired us
We were driven by how quickly misinformation spreads online—from viral text posts to manipulated images and deepfake videos. Traditional fact-checking is too slow, often arriving hours or days after the damage is done. What inspired us was the realization that people don’t just need a verdict like “true” or “false”—they need to see the reasoning behind it. We wanted to build a system that not only detects misinformation but also explains, educates, and empowers users to make informed decisions in real time.
What we set out to learn
We aimed to:
- Deploy explainable AI that doesn’t just flag misinformation but shows why something is likely false.
- Integrate community-driven validation to enhance trust and accountability.
- Offer multi-modality detection, handling text, images, video, and audio.
How we built it
We adopted an Agile, iterative prototyping model inspired by rapid prototyping approaches.
- Prototype 1: A Chrome extension that detects deepfake video and manipulated images using a Vision Transformer + LLM combo.
- Prototype 2: Backend APIs for cross-modality fact validation—extracting statements, querying trusted knowledge bases, and returning credibility scores with scanned source links.
- Prototype 3: A mobile/web interface that enriches AI flags with explainable visual cues (e.g., "mismatched facial landmarks," "text incongruence," "source discrepancy") and invites user votes to confirm/contest.
Challenges we faced
- Achieving real-time detection across modalities while maintaining accuracy and low latency.
- Collecting a reliable and diverse dataset of verified misinformation vs. legitimate content.
- Crafting clear, trustworthy explanations that are technically accurate yet understandable to non-experts.
- Designing a gamified community validation layer that balances speed, trust, and spam resistance.
The impact
- Acts as a truth-scanning lens for social media and live content.
- Empowers users with transparent insights, enabling better-informed decisions.
- Builds a hybrid trust pipeline: AI first, community validation next, with full traceability.
Built With
- docker
- fastapi
- github
- javascript
- jupyter-notebook
- numpy
- pandas
- python
- pytorch
- react
- roberta
- scikit-learn
- tailwind
- typescript
Log in or sign up for Devpost to join the conversation.