Towards Robust Fact-Checking: A Multi-Agent System with Advanced Evidence Retrieval

This repository provides scripts and workflows for translating fact-checking datasets and automating claim classification using large language models (LLMs).

Code associated with the NAACL 2025 paper "COVE: COntext and VEracity prediction for out-of-context images"

Tathya (तथ्य, "truth") is an Agentic fact-checking system that verifies claims using multiple sources including Google Search, DuckDuckGo, Wikidata, and news APIs. It provides structured analysis with confidence scores, detailed explanations, and transparent source attribution through a modern Streamlit interface and FastAPI backend.

Code associated with the preprint: "M4FC: a Multimodal, , Multilingual, Multicultural, Multitask real-world Fact-Checking Dataset"

debunkr.org Dashboard is a Browser extension that helps you analyze suspicious content on the web using AI-powered analysis. Simply highlight text on any website, right-click, and let our egalitarian AI analyze it for bias, manipulation, and power structures.

OpenSiteTrust is an open, explainable, and reusable website scoring ecosystem

🔍 ABCheckers 💬 is a data-driven project that analyzes Twitter discourse to uncover misinformation around 🇵🇭 inflation and the weakening peso, empowering users with contextual insights.

Media Literacy System powered by AI - Analyze news for bias and manipulation.

An advanced AI-powered fake news detection system that verifies text, images, and social media posts using Gemini AI, FastAPI, and Next.js. Includes a modern web interface, a lightweight Streamlit app, and a Chrome extension for real-time fake content detection. Built to combat misinformation with explainable AI results and contextual source links.

Fine-tuned roberta-base classifier on the LIAR dataset. Aaccepts multiple input types text, URLs, and PDFs and outputs a prediction with a confidence score. It also leverages google/flan-t5-base to generate explanations and uses an Agentic AI with LangGraph to orchestrate agents for planning, retrieval, execution, fallback, and reasoning.

This project implements a complete NLP pipeline for Persian tweets to classify topics and detect fake news. Using a Random Forest classifier, it compares tweet content with trusted news sources, achieving 70% accuracy in fake news detection.

A React + Vite + Tailwind CSS web app that verifies text for potential misinformation in real time using Gemini AI. Delivers a minimal, responsive UI with clear verdicts, confidence scores, and category tags. Includes a dashboard-ready structure and components for insights and community upvoting.

Trinetra AI is a creative solution brought to the market that produces verifications for AI-created text or other conceivable falsehoods. It supplies a quick and accessible web experience where users can receive the detected verdicts, confidence scores, and topical categories along with graphics of trend and community feedback.

Node.js + Express API that powers misinformation verification by integrating Gemini AI and MongoDB. Exposes endpoints for verification, category summaries, upvoting, and health checks, designed for low-latency responses. Persists flagged content with confidence and metadata for analytics and auditability.

Adventure Guardian AI is a unified safety intelligence system designed to protect adventure travellers in India. It verifies trek information, analyzes health risks, and detects fraud using AI-powered vision, geodata, weather intelligence, and pattern analysis. By combining truth, health, and fraud assessments, it generates a single Verified Trek S

Fact-checking Reddit posts with machine learning: comparing traditional and transformer-based approaches

Advanced NLP MLOps pipeline for misinformation detection, utilizing RoBERTa with LoRA (PEFT) for efficient fine-tuning. This project focuses on cross-domain generalization across the FakeNews-Kaggle and LIAR datasets, featuring robust data engineering, mixed-precision training, and comprehensive metric evaluation.

Imagine Hashing embeds cryptographic hashes into images using steganography and SHA256 to ensure authenticity, integrity, and resilience against tampering or manipulation.

AI-powered fake news detector with 3-tier verification: Google Fact Check API → Wikipedia → ML ensemble (90% accuracy). 4 models: Naive Bayes, Logistic Regression, Random Forest, SVM. NLP pipeline with spaCy NER, TF-IDF vectorization. Professional Flask web app with real-time analysis.