Science Fair Origin
Started as a high-school scientific fair project focused on the social problem of misinformation.
Award-Winning AI Project
FNDetector
AI-assisted fake news detection platform originally created for a 2025 high-school science fair and later evolved into a portfolio-ready Flask, NLP and machine learning application.
Local Recognition
Won first place at the original science fair where the project was presented.
International-Level Fair
Presented at MostraTEC, one of the largest science and technology fairs in Latin America.
Computer Science
Achieved third place in Computer Science, reinforcing the technical relevance of the project.
Overview
What is FNDetector?
FNDetector is an educational and applied artificial intelligence project designed to help users evaluate suspicious news claims. It combines a web interface, a Flask backend, local labeled data, BERT embeddings and an explainable legacy verification baseline.
The goal is not to replace professional fact-checking, but to provide an assistive signal that encourages users to investigate information before sharing it.
Detection Approaches
Two complementary analysis modes
BERT Classifier
Uses embeddings from the Portuguese BERT model neuralmind/bert-base-portuguese-cased with a trained logistic regression classifier to estimate whether a claim resembles true or fake examples from the dataset.
Legacy Verifier
Searches for related articles, extracts readable text, compares textual similarity and returns a transparent corroboration score based on retrieved sources.
Technical Stack
Technologies used
Python
Flask
BERT
Transformers
Scikit-Learn
Logistic Regression
NLP
HTML
CSS
JavaScript
Architecture
How the project is organized
FNDetector/
├── app.py
├── data/training/
├── docs/
├── models/
├── src/fndetector/
├── web/pages/
├── web/static/
├── requirements.txt
└── README.mdCore Features
What the project demonstrates
Flask web application with multiple user-facing pages.
BERT-based fake news classification pipeline.
Legacy similarity-based verification baseline.
Training-data upload endpoint for new samples.
Local documentation library served through stable routes.
Portfolio-ready project structure with separated modules.
Impact
Why this project matters
FNDetector connects artificial intelligence with a real social issue: misinformation. It was built not only as a technical experiment, but also as a project that could be presented to judges, students, teachers and the public.
Its evolution from a science fair prototype into a structured software repository shows both research communication and practical engineering skills.
Explore FNDetector on GitHub
View the source code, documentation, project structure and development history of FNDetector.
Open GitHub Repository