Fakeddit

A headline classifier with a separate source-research workflow and an experimental text-plus-image pipeline.

Repository browser capture of the Fakeddit workspace in its empty input state; no news verdict is shown.

Overview

The serving path uses an exported TF-IDF and logistic regression text model, including a portable inference implementation. It keeps dataset classification separate from credential-dependent source research. A distinct CLIP and XGBoost pipeline supports paired text-image experiments.

Features

• Portable TF-IDF / logistic inference with low-coverage abstention and uncertain-output handling.
• Separate Groq-backed claim research when configured.
• Independent text-image pairing, local CLIP embeddings and XGBoost training code.