CheckLah!

To combat online misinformation, we are developing an application that enables users to upload information to access their credibility.

Our research shows that, despite many tools detecting misinformation, they are few user-friendly applications analyzing multi-modal data.

Hence, our project aims to cover both text and image inputs to assess content authenticity. Text inputs undergo NLP analysis utilizing libraries like “sentiment,” “compromise,” and “text-statistics.” Additionally, the FB-BART-MNLI model (https://api-inference.huggingface.co/models/facebook/bart-large-mnli) from Hugging Face is used to evaluate content credibility through ML. This analysis is powered by a Node.js backend that we set up, hosted on Render.

For image inputs, OCR is employed to extract text if present. Otherwise, the image undergoes deepfake and AI detection using the Sightengine API (https://api.sightengine.com/1.0/check.json).

API keys are used to securely send requests to the aforementioned models, from which we fetch responses.

Processed results are displayed on the frontend in a simplified way, providing users with comprehensive insights into the content’s authenticity.

Our current development is a PWA, which focuses on the essential aspect. However, it could be more user-friendly by enabling CheckLah to appear on the share sheet, reminding responsible information dissemination, which aligns with our vision of inculcating a habit of checking before sharing information.