Climate Claim Fact Checking Agent with Evidence Retrieval

Can a small local language model fact-check climate claims if every output must be a validated object? This project builds a retrieve, judge, aggregate agent with Instructor on LiteLLM and a Chroma evidence store: each sentence gets a stance, and a verdict may only cite retrieved sentences. The model (qwen2.5:3b) runs in local Ollama on a laptop with RTX 2050 4 GB (Ryzen 7, 8 GB RAM).

NOTE: Automated assessment for research and teaching on the public Climate-FEVER claims. A small local model can be wrong: read the cited sentences before relying on a verdict. The app never publishes verdicts; a person decides.

Category: Department of Computer Science and Engineering / Agent Development

Agent Development Project AG17 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.