Crosscheck AI Copilot Flags Contradictions Between Research Sources Using Local LLMs

A developer has built Crosscheck, an AI research assistant layered on top of the Cognee knowledge graph framework, designed to detect when two sources report conflicting facts about the same subject. Unlike traditional graph-based approaches, Crosscheck extracts claims directly from raw source text as structured subject-predicate-object triples, preserving exact values and tagging them with source IDs and timestamps. A two-stage engine first filters candidate contradictions structurally, then uses an LLM to confirm whether the conflicting claims cannot both be true. The tool runs fully offline via Ollama with models like Llama 3.1 8B, and all ingested data persists so the copilot can answer queries without re-processing documents. The same contradiction-detection logic was also applied to financial auditing in a companion tool called Argus, which identified $5,300 in contract and invoice discrepancies across a demo dataset.