AI Fake News Detection System

I’m creating a production-ready AI/ML solution that automatically reviews articles pulled from news websites and flags any item that is likely to be fake. All content will be in English, and the system must consistently reach at least 90 % accuracy on an independent validation set; precision and recall should be reported alongside overall accuracy so I can see how it performs on both real and fabricated pieces.

The workflow I have in mind combines three parts:
• an automated pipeline that scrapes or ingests fresh articles, cleans the text, and stores it in a structured format;
• a robust NLP model—think transformer-based architecture fine-tuned on a well-curated fake-vs-real news dataset—trained and evaluated in Python with libraries such as PyTorch, TensorFlow or Scikit-learn;
• a lightweight interface (REST API or simple web demo) that returns the probability of falseness plus the key features that influenced the decision for transparency.

Deliverables
• End-to-end codebase with clear instructions (requirements.txt / environment.yml, README)
• Trained model weights and scripts to reproduce training and evaluation
• Detailed report showing dataset splits, confusion matrix, precision, recall, F1 and overall accuracy ≥ 90 %
• Deployment script or Dockerfile so I can run the service on my own server

Acceptance criteria
1. Running the provided setup command spins up the API or web demo without errors.
2. Inference on a supplied test set of English news articles matches or exceeds the promised performance metrics.
3. Code is commented, modular and adheres to standard Python style so future contributors can extend the project easily.

If you already have experience fine-tuning BERT-style models for misinformation detection or have worked with fact-checking datasets, that will help us move quickly. I’m ready to review initial architecture ideas and sample outputs as soon as you have them.