AIC CTU@AVerImaTeC: dual-retriever RAG for image-text fact checking
A simple, three-module pipeline using GPT-5.1 took 3rd place in a major AI verification challenge.
Deep Dive
Researchers Herbert Ullrich and Jan Drchal from CTU built 'AIC CTU@AVerImaTeC', a dual-retriever RAG system for image-text fact-checking. It combines a text similarity search, a reverse image search API, and GPT-5.1 for generation. The system placed 3rd in the AVerImaTeC shared task, costs only $0.013 per fact-check, and its modular code is published for easy reproduction and further experimentation by developers.
Why It Matters
Provides a cost-effective, open-source blueprint for building reliable AI systems to combat visual misinformation at scale.