Fake News Detector

The Challenge

Misinformation spreads quickly, and flagging it takes more than a keyword list. The backend had to:

• Detect harmful content: Flag untrustworthy articles with a machine learning model, not just heuristics
• Answer in real time: Score an article while the user waits
• Stay reliable: Keep working when a model or an input misbehaves
• Deploy anywhere: Run the same way in development and production

The Approach

I engineered a backend pipeline in Python and Flask that detects and flags harmful or untrustworthy content, with an ML model at its core.

ML-Backed Credibility Analysis

• Integrated a RoBERTa model (TensorFlow) for real-time credibility analysis of each article
• Combined it with TextBlob sentiment analysis across 5 weighted scoring factors
• Added heuristic-based fallbacks so the pipeline still returns a score when the model can't

Backend Performance

• Loaded the ML model once with the Singleton pattern, cutting per-request overhead while keeping detection reliable

Containerized Deployment

• Architected a containerized stack with Docker Compose and multi-stage builds
• Hot-reload containers for development, production-ready images for deployment

Dashboard

• A React dashboard shows each article's credibility score and the factors behind it

Technical Highlights

• Python and Flask detection pipeline
• RoBERTa (TensorFlow) model for credibility analysis
• TextBlob sentiment as one of 5 weighted factors, with heuristic fallbacks
• Singleton model loading for lower per-request overhead
• Docker Compose with multi-stage builds