Inspiration
Misinformation spreads rapidly online, leading to confusion and even harm. We wanted to create a solution that makes fact-checking as seamless as sending a message, empowering people to verify content effortlessly.
What it does
"real_anot" is a Telegram bot that verifies the authenticity of text messages and videos. Whether it’s scam alerts, misleading headlines, or AI-generated deepfake content, "real anot" offers an interactive, and entertaining way to separate truth from falsehood, ensuring fact-checking is always just a message away.
How we built it
For text verification, we employed a Large Language Model (LLM), specifically OpenAI’s GPT API, to assess whether it is falsehoods and provide concise explanations. If misinformation is detected, the bot doesn’t just provide a fact-checked explanation, but it also generates a hilarious meme, making the debunking process fun and highly shareable. For video verification, we leveraged on Hugging Face’s deepfake detection models, which identify AI-generated or manipulated content.
Challenges we ran into
Some of the key challenges we’ve encountered with "real anot" include:
Accuracy of AI Models: While the APIs for text and video verification are powerful, they aren’t perfect. False positives or negatives can sometimes happen, which could lead to trust issues from users if they don’t get the correct results.
Integration with Telegram: Integrating telegram bot with APIs was not easy as it required us to understand and implement codes specific to the different technologies.
Large requests: When dealing with large video files or managing a high volume of simultaneous requests, there would be slow responses.
Accomplishments that we're proud of
We created a working prototype that quickly and accurately verifies content, wrapped in a user-friendly interface. The meme generation feature added a layer of engagement we’re really excited about!
What we learned
We gained valuable experience in working with AI models for content verification, optimizing APIs for performance, and designing a frictionless chatbot experience within Telegram.
What's next for real_anot
We plan to expand support for more refine deepfake detection, and implement a more diverse check for different types of fake news sources, such as through URLs and images. We also want to run the bot on a cloud platform so it remains active even when our computers are offline.
Log in or sign up for Devpost to join the conversation.