-
-
AI-powered system that generates and detects fake news using advanced NLP and machine learning.
Inspiration: With misinformation spreading faster than facts, we wanted to create a tool that helps people understand both how fake news is generated and how it can be detected using modern AI. This inspired us to combine Generative AI and Transformer-based detection in one platform.
What it does: Generates fake news using GPT-2
Detects fake vs. real news using BERT/DistilBERT Explains predictions with LIME, highlighting important words and showing confidence scores
How we built it: Cleaned and prepared a real fake-news dataset
Fine-tuned BERT/DistilBERT for classification using GPU Integrated GPT-2 with safe, controlled generation settings Added an explainability layer with LIME Built a Flask API backend and Streamlit UI for smooth user interaction
Challenges we ran into: Managing noisy, imbalanced datasets
Limited GPU time for fine-tuning Keeping GPT-2 outputs safe and non-harmful Integrating explainability with deep learning models
Accomplishments that we're proud of: Complete end-to-end system: Generation + Detection + Explainability
High-accuracy transformer model Clean, user-friendly interface Scalable and modular architecture
What we learned: Effective fine-tuning of transformer models
Importance of explainable AI Safe generative AI techniques Building and deploying full-stack ML applications
What's next for AI-Powered Fake News Generator & Detector (Pro Version): Deploy to cloud (HuggingFace Spaces / Render / Streamlit Cloud).
Add multilingual detection (Hindi, Bengali, Japanese, etc.). Train on larger global news datasets. Build a browser extension for real-time news credibility checks. Integrate RLHF for more responsible text generation. Add fact-checking APIs using trusted sources.
Log in or sign up for Devpost to join the conversation.