Grounding snippets are query-specific extractive summaries passed to an LLM re-ranker to ground responses in live search facts, which DEJAN optimizes through substantive micro-edits that satisfy the model's ranking preferences.

flowchart LR
    A[User Query] --> B[Search Index]
    B --> C[Extractive Snippets]
    C --> D[LLM Re-ranker]
    D --> E[Final Grounded Answer]