DEJAN influences retrieval-grounded search by parsing live LLM citation mechanics and using a Bayesian Content Optimizer to iteratively rewrite web page grounding snippets until the model ranks them as the preferred source.
flowchart LR
A[Web Page Content] --> B[Snippet Generator]
B --> C[Candidate Snippet]
C --> D[LLM Grounding Probe]
D --> E[Citation Parser]
E -->|Bayesian Optimization| B