member models in the ensemble model
how it works
tech stack
credible data sources
data pipeline
frontend retro-style design
more explicit pipeline
example run 1
example run 2

MYTH CHASER: Fighting Misinformation with AI

🎯 What Inspired This Project

Watching family members fall for WhatsApp scams and misinformation made us realize we need an accessible AI tool to instantly verify suspicious content. In our post-truth era, distinguishing fact from fiction shouldn't require a journalism degree. And thus, this project is the one that caught our attention the most out of the three proposed ones.

🧠 What We Learned

Ensemble AI: Combining 7 different models yields more accurate results than any single approach.
Multimodal Processing: Each content type (text, images, audio) requires specialized analysis techniques.
Confidence Scoring: Developed weighted voting algorithms to aggregate conflicting AI predictions into meaningful truth scores.

🔧 How We Built It

Architecture

Microservices architecture with an AI ensemble voting system.

Pipeline

User Input (Text/Files) ➜ Language Detection ➜ Content Extraction ➜ Context Building ➜ Parallel Processing Ensemble model ➜ Verdict Determination (FACT/MYTH/SCAM) ➜ Evidence Collection ➜ Response Formatting ➜ API Response

Tech Stack

Backend: Python, FastAPI
AI/ML: PyTorch, HuggingFace Transformers, Groq
Frontend: Next.Js

Models Used

Google Fact Check API
ClaimBuster
Custom SBERT-based semantic similarity model
NLI (Natural Language Inference) Transformer
Fake News Classifier (custom-trained)
TuniBERT
Groq

🚧 Key Challenges

Model Disagreement: Solved with weighted voting based on each model's domain expertise.
Context Understanding: Added OCR and cultural context preprocessing.
Performance: Enabled parallel processing and smart model selection to reduce latency.