-
-
cover page
-
game page
-
misinformation checker page
-
chatbot page
🚀 Inspiration
In today’s digital world, misinformation spreads rapidly, making it challenging to distinguish between real and manipulated content. Inspired by the increasing prevalence of AI-generated deepfakes and fake news, we created Debunkd—a platform that empowers users to verify images and test their fact-checking skills in an engaging way.
We wanted to combine gamification and AI-powered analysis to create a tool that is both educational and fun while leveraging cutting-edge LLMs, OCR, and fact-checking APIs. Our goal was to make misinformation detection accessible to everyone, whether they are playing the game or using AI-powered tools for verification.
🔍 What It Does
Debunkd provides two main features. The first is The Game, where users are presented with an image and must decide whether it’s real or fake. A score tracker keeps track of their performance, and high scores are recorded to encourage competitive play. The second feature is AI-Powered Image Verification, which allows users to upload an image (or select a sample) for analysis. The system processes the image using OCR, RAG-based fact-checking, and Google Fact Check API to determine its authenticity.
Additionally, Debunkd includes a chatbot with memory retention, enabling users to ask follow-up questions for deeper insights. The chatbot provides relevant, context-aware responses instead of generic answers, making the fact-checking experience more interactive and informative.
🛠️ How We Built It
The frontend of Debunkd is built using HTML, CSS, and JavaScript, creating a simple yet engaging user experience. The backend is powered by Python (Flask), handling API requests for image processing and fact verification.
For AI and image processing, we integrated multiple technologies. Tesseract OCR extracts text from images, while LLM (ChatGroq) and LangChain process and evaluate misinformation. ChromaDB enables vector-based similarity search for retrieving relevant fact-checking data. Additionally, we integrated Google’s Fact Check API to cross-reference claims with real-world verified information.
Our database is powered by Supabase, storing game scores and a dataset of real/fake images. For deployment, we used Vercel for the frontend and Render for the backend, ensuring scalability and smooth user interactions.
⚠️ Challenges We Ran Into
One of the biggest challenges was extracting accurate text from images. OCR sometimes struggled with low-quality images or distorted text, making it difficult to analyze content properly. Another challenge was LLM hallucinations, where the AI would occasionally provide misleading responses. To overcome this, we refined the prompts and optimized our retrieval-augmented generation (RAG) pipeline.
Additionally, the fact-checking process was initially slow due to searching large datasets. We improved response time by optimizing API calls, caching results, and refining the vector search parameters in ChromaDB. Finally, designing a balanced and engaging game mode was a challenge, as we needed to ensure that real and fake images were challenging enough without making the game too easy or frustrating.
🏆 Accomplishments That We're Proud Of
We are proud of successfully integrating multiple AI technologies into a single, functional system. We built an interactive game mode with real-time score tracking, allowing users to learn and test their misinformation detection skills in a fun way. Additionally, our chatbot with memory retention enhances user interactions, making fact-checking more intuitive.
Another key accomplishment was optimizing image verification speed while maintaining high accuracy. By refining our OCR and retrieval-based fact-checking, we created a system that efficiently detects misinformation from images and text.
📚 What We Learned
Throughout this project, we learned that fine-tuning LLM prompts significantly improves response accuracy and reduces misinformation hallucinations. We also discovered that ChromaDB is a powerful tool for retrieval-based fact-checking when optimized correctly.
Additionally, we found that gamification is an effective strategy for engaging users in misinformation detection. By making the learning process interactive, users are more likely to develop critical thinking skills in evaluating media content. Finally, optimizing API calls and caching data proved crucial in improving overall system performance.
🚀 What's Next for Debunkd
In the future, we plan to introduce a multiplayer mode, allowing users to compete in real-time fact-checking challenges with friends. We also aim to integrate deepfake detection for video verification, expanding Debunkd’s capabilities beyond static images.
Another feature in development is user-generated challenges, where users can submit images for the community to verify. Additionally, we are exploring the possibility of launching a mobile app version of Debunkd, making AI-powered fact-checking accessible on iOS and Android devices.
Our mission is to continue refining Debunkd to help users combat misinformation effectively, whether through gaming, AI-powered verification, or interactive chatbot assistance.
Log in or sign up for Devpost to join the conversation.