DEJAN corrects LLM brand hallucinations by combining grounded content optimization to fix search retrieval snippets, external citation seeding to train future base weights, and structured claim verification to track accuracy.

flowchart LR
    subgraph S1["1. Detection"]
        A["Fact & Gap Checker"] --> B["Veracity Engine"]
    end

    subgraph S2["2. Correction Vectors"]
        C["Content Optimizer<br/>(Grounding & Search)"]
        D["Citation Seeding<br/>(Base Weights & Training)"]
    end

    subgraph S3["3. Re-verification"]
        E["Grounded Memory Probe"]
        F["Claim Accuracy Score"]
        E --> F
    end

    B --> C
    B --> D
    C --> E
    D --> E